swisper-agent-sdk

Build a Swisper domain agent.

This package is self-contained. You do not need the Swisper backend, a database, or network access to build an agent and test it.

Release 0.1.0a30. It succeeds 0.1.0a19, the previous published release. Between releases this source page follows the development tree, so the immutable published package is not the SDK this page describes as later changes land. Pin the published version rather than assuming a source checkout and an installed release contain the same changes.

Build against the documented surface — Agent, Delegation, Reply, d.recall, d.generate, d.card, d.say, harness and AgentContract. Anything not documented on this page may move without a deprecation cycle. Pin swisper-agent-sdk==0.1.0a30 rather than a range while releases are alpha versions, and tell us you have pinned it — we will tell you before a change that would break you rather than after. There is no deprecation policy yet because there is not yet a partner whose upgrade we would be managing; when there is, that is the moment it gets written, and you will be told rather than left to read it here.

Published a19 and current source documentation

0.1.0a19 was published from reviewed source 9ced89117b9eeacd438fd7c7b42d6ac9532de250, tagged agent-sdk-v0.1.0a19. The wheel and sdist were downloaded and matched against their GitHub Release assets, reviewed logical contents and cryptographically verified publication attestations. This release includes complete registration metadata transmission and the repair that preserves the published a18 positional Delegation call sequence.

The README embedded in the immutable a19 wheel/sdist and its released PyPI description still contain build-time references to a18 as the install pin and to a19 as unpublished. Those frozen references are superseded by the current installation advice below; the released artifacts have not been rewritten or republished. This correction updates active source documentation only.

SDK publication does not establish a live provider connection, provider mapping or device action. Those require their own normal authorization and provider evidence.

Dated release history — a18, 2026-09-13 UTC

0.1.0a18 first published the following features, which were absent from 0.1.0a17 and earlier releases:

• The prompt-readback path: http_prompts_transport, HttpPromptsTransport, PromptsUnavailable, CAPABILITY_PROMPTS_PATH and PromptLibrary.from_core.
• The delegation-token path: current_delegation_token, delegation_token_from_body and use_delegation_token.
• Additional tracing exports: setup_tracing, adopt_trace_context, TraceTree, Observation and note_degradation. TraceContextMiddleware was already published.
• Typed refusals: CapabilityRefused, DelegationTokenRefused and ProviderCallFailed.
• The doctor CLI command, and the conformance and tool_envelope modules.

The downloaded swisper_agent_sdk-0.1.0a18-py3-none-any.whl was inspected by AST: 115 exports and 28 top-level modules, matching the tree at tagged commit c1ba8c7af. The earlier comparison against a17 found 100 exports and 26 modules. These are dated a18 measurements, not counts for the current release.

Published a18 and earlier artifacts remain immutable. The historical evidence above is retained; the current installation pin is a30.

Install

Requires Python 3.11 or newer.

python3 -m venv .venv
./.venv/bin/python -m pip install "swisper-agent-sdk==0.1.0a30"

Use the virtual environment's interpreter for everything below. Typing python3 rather than ./.venv/bin/python is the most common way to get a ModuleNotFoundError that looks like a broken package and is not.

A handful of declared dependencies pull in far more packages. This package declares pydantic, langgraph, jsonpath-ng, pyyaml, langchain-core, httpx, jsonschema, swisper-studio-sdk, redis, jinja2 and typing-extensions — and langgraph alone brings its own dependency tree with it (langsmith, orjson, ormsgpack, zstandard and more). Measured on 2026-09-12 by installing 0.1.0a17 into a clean virtual environment on Python 3.14.7: 66 packages, not eleven. (The same install measured 62 on 2026-08-28 — the package did not change; its upstream tree did.) None of that is a defect — every one of them is something a declared dependency genuinely needs — but a partner watching 66 packages scroll by after being told about eleven deserves to have been told first.

🔴 This paragraph has now under-counted twice, and the second time it was the paragraph whose entire subject is the first time. It said 41 while 62 installed — a cold reader measured it. And then 62 went stale on its own, with nobody touching this package: a third reader measured 66 on 2026-09-12. So treat every count on this page as a dated observation, not a guarantee: they move when upstream changes, with no edit here and no release of ours. Re-derive rather than cite:

python -m venv /tmp/probe && /tmp/probe/bin/pip install swisper-agent-sdk
/tmp/probe/bin/pip list --format=freeze | wc -l

One of those 66 is langsmith, and importing this package switches it OFF. langsmith is LangChain's own tracing client. It is not optional and cannot be removed: langgraph requires langchain-core, which hard-requires langsmith, and langchain-core publishes no extras to opt out of. So it installs whether or not anyone wants it.

It exports by reading an environment variable at call time, with no involvement from any code here — and those variable names say nothing about this package, so a shared CI block copied between repositories can arm it for an agent whose author never heard of it. What would leave the process is the contents of the turn.

🔴 So import swisper_agent_sdk sets LANGCHAIN_TRACING_V2, LANGCHAIN_TRACING, LANGSMITH_TRACING_V2 and LANGSMITH_TRACING to false in the host process, and sets langsmith's own global tracing switch to off. This is the one thing importing this package changes outside itself, and it is documented here because it is a change to your process, not ours: if your service legitimately uses LangSmith, this turns it off, and you will want swisper_agent_sdk.langsmith_off in view before you debug why.

Re-enable it after import if you mean to. Observability for agents built with this SDK goes to Studio instead — see the tracing section.

Both numbers move as the package evolves, so read them from the source rather than from this paragraph — the declared list has changed four times already:

python3 -c "import importlib.metadata as m;print(*[r for r in m.requires('swisper-agent-sdk') if 'extra ==' not in r],sep=chr(10))"

That reads the installed package, so it works however you installed it. (An earlier version of this line read pyproject.toml off the disk — which raises FileNotFoundError after pip install swisper-agent-sdk, i.e. for exactly the path this page recommends first. A cold reader hit it. The one command offered here as more trustworthy than the prose has to work for the reader who runs it.)

Not every declaration costs a download. Measured by resolving the manifest with each one removed: httpx costs 0 packages — langchain-core, langgraph-sdk and langsmith already require it — while jsonschema costs 5, since it brings attrs, referencing, rpds-py and jsonschema-specifications with it. jsonschema earns that: it validates card payloads against the schemas your agent declares, at the boundary, so a malformed card is refused where it is produced rather than somewhere downstream. httpx is declared despite costing nothing, because an import we do not declare breaks the day an upstream drops it — with an error naming a package our own manifest never mentions.

From a directory, instead

Handed the package as a tarball or a source checkout rather than installing from PyPI? Point pip at the directory instead of the package name — everything else on this page works exactly the same either way:

python3 -m venv .venv
./.venv/bin/pip install -e ./swisper-agent-sdk

Your first agent

This is swisper_agent_sdk/examples/twelve_line_agent.py, complete and unabridged. It ships inside the package itself, so installing it is the only step — there is nothing separate to fetch:

from swisper_agent_sdk import Agent, Delegation, Reply

agent = Agent(
    name="echo_agent",
    description="Repeats the task back. The smallest thing that is still an agent.",
)


@agent.handler
async def handle(d: Delegation) -> Reply:
    if not d.task:
        return Reply.ask("What would you like me to do?")
    return Reply(text=f"You asked me to: {d.task}")

Run it — the same command whether you installed from PyPI or from a directory:

$ ./.venv/bin/python -m swisper_agent_sdk.examples.twelve_line_agent
You asked me to: book a train to Bern

d.task is what the user asked for. Returning a Reply answers them; returning Reply.ask(...) puts a question to them and pauses the turn.

All your agents in one place: My agents

swisper-agent ui opens the developer page. Run it in a folder that holds no agent (for example ~/Swisper Agents itself) and the page opens on My agents: one card for each agent in ~/Swisper Agents (--agents-dir points it elsewhere). Each card shows who built the agent and what Swisper says about it: Not submitted from here, Waiting for an operator, Going live, Live, Rejected, Blocked by Swisper or Suspended. Open a card to land on that agent's Start here. New agent creates the next one. The first time, the page asks you to register once on this machine; after that, no agent asks again.

Run swisper-agent ui inside an agent's folder and the page opens straight on that agent's Start here, as before. The header's ‹ My agents and its switcher move between agents.

What always arrives, and what you ask for

Eight fields reach your agent on every delegation, whatever it declares. They are the turn's context — who is asking, which conversation this is, and how to speak back. You read them from d.raw:

field	what it is for
chat_id	The conversation this turn belongs to. Key anything you keep between turns on it.
user_id	Who the turn is for. Use it to scope anything you store yourself.
user_input	What the user actually typed, verbatim. d.task is the supervisor's instruction derived from it — when you need their exact words rather than the instruction, this is the field.
user_locale	Regional formatting, e.g. de-CH — dates, numbers, currency.
user_timezone	e.g. Europe/Zurich. Anything you resolve against "today" or "tomorrow" needs it.
current_time	The turn's own clock, ISO-8601, from the user's device. Use this rather than your process's clock: your process may be in a different timezone, or on a different day.
interaction_language	ISO 639-1, for user-facing text. Detected upstream and recomputed on every turn and every answer to a question. When it is set, write your reply in that language rather than detecting it yourself.
llm_reasoning_language	The language the model reasons in, which is not necessarily the one it answers in.

Everything else is declared. A field outside these eight reaches d.raw only if your agent asks for it by name:

agent = Agent(name="...", description="...", inputs=("resolved_entities",))

That is the whole rule: core plus declared. A field you did not declare, and that is not one of the eight, is not in d.raw at all — not None, absent. So prefer d.raw.get("...") over d.raw["..."] for anything you declared conditionally.

These eight are what your code reads. Four of them — language, locale, timezone and the current time — are also composed into the model's prompt for you: d.generate builds them into a system message and prepends it to every call, ahead of the messages you passed. So you rarely need to restate them in instructions=; see PROMPTING.md, "Context you get for free". Reading interaction_language yourself is for branching in code, not for telling the model which language to answer in — that part is already done.

d.task is none of the eight, and needs no declaring

d.task comes from the envelope's current_plan, and the facade hands it to you whether or not you declare anything. The consequence is worth knowing before it puzzles you: for an agent that declares nothing, "current_plan" in d.raw is False while d.task is populated. Use d.task; it is the supported way to read the instruction.

d.task survives a resume

When your agent returns Reply.ask(...), the turn pauses and resumes later with the user's answer in d.answer and d.resumed set to True. d.task still holds the original instruction. It is not replaced by the answer, and you do not have to stash it yourself:

@agent.handler
async def handle(d: Delegation) -> Reply:
    if not d.resumed:
        return Reply.ask("Which day?")
    # d.task is still "book me a train"; d.answer is "Thursday"
    return Reply(text=f"{d.task} — on {d.answer}")

To carry anything else across that pause, hand it to Reply.ask(..., resume_with=...) and read it back from d.resume_with on the resuming turn. That is the supported bookmark; reaching into the raw envelope's HITL state instead is not.

Running a turn yourself

harness runs a real turn against your agent with no core, no database and no network:

import asyncio
from swisper_agent_sdk import harness
from my_agent import agent

result = asyncio.run(harness.turn(agent, "book a train to Bern"))
print(result.text)

Check your install

Needs the directory install above — tests/ ships in a source checkout, not in the plain PyPI wheel:

$ cd swisper-agent-sdk
$ ../.venv/bin/pip install -e ".[dev]"
$ ../.venv/bin/python -m pytest

That runs a handful of smoke tests: the package imports, importing it leaves your logging alone, an agent answers a turn offline, and every bundled example runs and prints something. If those pass, the SDK works on your machine.

Run pytest from inside this directory. The package's own asyncio_mode = "auto" setting lives in its pyproject.toml, and pytest only picks it up when invoked from here. Run it from the parent directory and every async def test fails — including correct ones — which looks like a broken SDK and is not.

[dev] is what installs the test toolchain. The base install deliberately does not: it is the runtime, not the toolchain.

It is pytest and pytest-asyncio that you need, but [dev] installs more than that — measured on 2026-09-12 against 0.1.0a17 on Python 3.14.7, [dev] takes a clean install from 66 packages to 76, adding exactly ten: pytest, pytest-asyncio, mypy, mypy_extensions, ruff, pluggy, iniconfig, pathspec, ast_serialize and librt. (On 2026-08-28 the same two numbers were 62 and 72, and the added set included Pygments, which no longer appears — the delta is stable at ten, its membership is not.) Those are our tools for developing the SDK, and you inherit them. If you would rather not, pip install pytest pytest-asyncio alongside the base install is enough for everything on this page. (Recorded because the paragraph above warns you about exactly this under-counting for the base dependencies, and then this line did it again.)

Writing your own tests

🔴 Do this first or every test below fails. Put an asyncio_mode setting in your own project — the SDK's own setting does not reach your test files, and without it pytest refuses every async def test with "async functions are not natively supported", which names neither this SDK nor the fix:

# pytest.ini, next to your tests
[pytest]
asyncio_mode = auto

_{Or, in pyproject.toml: [tool.pytest.ini_options] / asyncio_mode = "auto". You also need pytest and pytest-asyncio installed — either pip install pytest pytest-asyncio, or pip install swisper-agent-sdk[dev], which brings those two plus our own linting toolchain (10 packages on top of the base install; see "Check your install"). This paragraph exists because a developer reading this page cold copied the example below verbatim and got four failures out of four; the requirement was documented, but under "Check your install", where it read as being about the SDK's own test suite.}

turn() returns a TurnResult with .text, .status, .resume_with, .cards, .said, and two helpers for assertions — .said_something_about(...) and .asked_for(...):

async def test_it_answers():
    r = await harness.turn(agent, "weather in Bern tomorrow")
    assert r.said_something_about("Bern")


async def test_it_can_ask_a_question():
    r = await harness.turn(agent, "book me a train")
    assert r.asked_for("departure time")
    r = await harness.answer(r, "09:00")     # resumes the paused turn

harness.answer(...) resumes a turn your agent paused with Reply.ask(...). Inside the handler, d.answer is what the user replied and d.resumed is True.

What a green harness run does and does not prove. The harness stands in for the supervisor — it builds the delegation a real one would send — and no token is minted or verified anywhere on this path. It does not stand in for the inference capability: d.generate reaches whatever generate_transport= you gave the agent, so a green run proves your agent's logic against your stand-in, not that the capability works against a live core. Those are different claims and only the first one is tested here.

Saying things, and sending cards

Two ways to put something in front of the user besides your final answer.

await d.say("…") narrates a deterministic moment — typically just before a slow call, where waiting for the model to decide to mention it is the wrong behaviour:

await d.say("Checking which trains actually run at that hour…")

It publishes onto the same channel Swisper republishes to the frontend, marked as a process indicator rather than answer text — so it is shown as a status line and is not concatenated into the reply the user keeps. It also stays in result.said for your tests.

await d.card("swi-transport", payload) emits a card. The types your agent may emit are declared on the agent, and an undeclared type raises UndeclaredCardTypeError — never a silent drop:

agent = Agent(name="…", description="…", cards=["swi-transport"])

await d.card("swi-transport", {"from": "Zurich", "to": "Bern"})   # fine
await d.card("swi-invented", {...})                               # UndeclaredCardTypeError

The declared list is also what the prompt tells the model about, so it cannot invent a card type that has no renderer. Both are visible in tests as result.said and result.cards.

No mapper, no registry — and you do not need to read card_builder.py. d.card() serialises the dict you give it directly; that module is Swisper's own internal rendering pipeline for its own cards, and none of it is something a partner calls. (It is importable from the package root, has "card" in the name, and is 660 lines long — a cold reader read all of it before finding this out. Hence this paragraph.)

If you call it anyway, know that it spells the type differently. d.card() takes the type exactly as you declared it, prefix included, and writes it into the tag verbatim: d.card("swi-transport", …) → <swi-transport …>. build_card_xml() takes the bare name and prepends swi- itself: build_card_xml("transport", …) → <swi-transport …>, while build_card_xml("swi-transport", …) → <swi-swi-transport …>, which nothing renders, with no mistake of your own. Two entry points, two spellings; the one this page teaches is d.card()'s.

You can have your payloads validated, and you should. Declare a JSON Schema per card type and the SDK checks every payload against it, at the point of emission:

agent = Agent(
    name="…", description="…",
    cards=["swi-transport"],
    card_schemas={
        "swi-transport": {
            "type": "object",
            "required": ["from", "to"],
            "properties": {"from": {"type": "string"}, "to": {"type": "string"}},
        }
    },
)

await d.card("swi-transport", {"from": "Zurich"})   # CardPayloadInvalidError

The error names the card type, the failing JSON-Path and the failing keyword, so a malformed payload is refused where it is produced rather than surfacing as a blank space in someone's chat. card_schemas is optional — omit it and payloads are passed through unchecked — but it is the difference between finding a bug in your own test run and finding it in a user's conversation. This is why jsonschema is a declared dependency and the only one that costs a real download.

What d.card() does to your reply text

🔴 Worth knowing before it surprises you. An emitted card is prepended to the reply text as a hidden XML block, so Reply.text — and result.text in the harness — is not only the sentence you wrote:

<swi-transport style="display:none" version="1.0.0" data-format="json">
{"from": "Zurich", "to": "Bern"}
</swi-transport>

Your train leaves at 08:04.

The frontend parses that block out and renders the card; the display:none keeps it invisible if anything ever renders the string raw. For assertions, prefer result.cards and result.said_something_about(...) over an equality check on result.text — an exact-match test will fail the moment the agent emits a card, and the diff will be dominated by markup rather than by the thing you were testing.

Who builds a card: you provide the data, Swisper builds the rendering

This is a division of labour, and it is deliberate.

	who
Deciding a card is the right way to show this	you, with us
The card's type name and the shape of its payload	you — you know your domain
Producing that payload at runtime, per turn	you, via d.card(type, payload)
The rendering pipeline — the component that turns your payload into something a user sees, its layout, states, theming, responsive behaviour and accessibility	the Swisper design team

You are not expected to build, style, or register a renderer, and there is no hook here for you to do so. Send us the card type you need and an example payload, and the design team builds the rendering for it. That is the whole process.

The practical consequence, so it does not surprise you mid-build: a card type whose rendering has not been built yet does not display. It is not an error and nothing raises — the payload travels correctly all the way to the frontend and simply has nothing to draw it. So agree your card types with us early, at the point you are designing the agent, rather than discovering it when you first see a real conversation.

UndeclaredCardTypeError cannot warn you about this, and it is worth knowing why: cards=[...] is checked against your own declaration, so it catches a typo in your code. Whether a renderer exists is a fact about the Swisper frontend, which the SDK never sees. In the same way, harness.turn() reports cards: ['your_type'] for any type you declared — that tells you your agent emitted it correctly, which is the half you own, and it is silent about the half we own.

Types that render today are all swi--prefixed — swi-transport, swi-transport-booking, swi-calendar, swi-email, swi-news, swi-places, swi-product, swi-reply, plus the payment and confirmation variants. If one of those already fits your data, use it and no design work is needed.

_{meal_plan in examples/meal_planning_agent.py is an illustration of the mechanism, not a type with a renderer behind it. While a card of yours is waiting on design, d.say(...) and your final answer text always reach the user.}

What is yours and what is ours

swisper_agent_sdk exports 125 names, and this page describes about twenty of them. That is not an omission — most of the rest are machinery the package needs across its own module boundaries, importable because Python has no other way to share them. You are not missing a chapter. But "is the thing I need in here and undocumented, or genuinely absent?" is a fair question with no answer until now, so:

module	what it is	yours?
agent	Agent, AgentGraphState	✅ yes — start here
delegation	Delegation and the errors it raises (CapabilityTimeout, CardPayloadInvalidError, CoreTooOldError, tier and entitlement errors)	✅ yes
transport	http_generate_transport — how your agent reaches a model through core; http_prompts_transport — how it reads its prompts from core (D-27)	✅ yes
contract	AgentContract, load_agent, and the declaration errors	✅ yes, if you use agent.contract.yaml
prompt_library	PromptLibrary — from_contract for local dev, from_core in production (reads what an operator edited, falls back to your bundled copy and says so)	✅ yes
nodes	BaseNode, MCPToolGroup, the EventSink protocol	⚠️ advanced — only for a multi-node graph. A single @agent.handler never needs them
state_types	UserInTheLoop, ToolOperation, ContextFile, CardMapping	⚠️ read-only types you may receive; you do not construct them
result_contract	DataObject, AgentResultEnvelope, the type registry	⚠️ advanced — typed results
card_builder	CardMapperRegistry, build_card_xml, attach_cards_to_text …	🚫 ours. Swisper's own card rendering pipeline. d.card() does not go through it — see above. ⚠ Takes the bare type (transport) and prepends swi- itself; d.card() takes the prefixed one
prompt_context, prompt_policy	prompt-assembly helpers	🚫 ours. d.generate already applies these for you; see PROMPTING.md
tracing	TracingConfig, TraceContextMiddleware, InMemoryTraceStore	✅ yes — and TraceContextMiddleware is one you must actually add; see below
domain_agent_interface, memory, state_keys, errors, time_utils	internal plumbing	🚫 ours

The rule, if you would rather have one than a table: anything this README names is yours and will be kept working. Anything it does not name is ours, and may move. Reaching into a 🚫 module is not forbidden and nothing will stop you — but it is not a supported surface, and we will not know we broke you.

If you need something you cannot find here, that is worth telling us: it is either missing, or documented badly, and both are ours to fix.

A fuller example

swisper_agent_sdk/examples/meal_planning_agent.py is the one to read once the twelve-line agent makes sense. It is a working agent that recalls what it knows about you, asks a question when it does not know enough, resumes on your answer, and emits a card:

$ ./.venv/bin/python -m swisper_agent_sdk.examples.meal_planning_agent
[turn 1] status=waiting_for_input text="I don't have any dietary restrictions or allergies on file for you — could you tell me what I should plan around?"
[turn 2] status=complete text='<meal_plan style="display:none" version="1.0.0" data-format="json">\n{"basis":"I\'m vegetarian and allergic to peanuts.", ...}\n</meal_plan>\n\nHere\'s a plan built around: I\'m vegetarian and allergic to peanuts..'
cards emitted: ['meal_plan']
plan basis: "I'm vegetarian and allergic to peanuts."

It also shows how to reach data you are entitled to. Declare what you need on the agent and read it from the delegation:

agent = Agent(..., inputs=("fact_lookup_service",))

# in the handler:
facts = await d.fact_lookup_service.get_by_type("ALLERGY")

You only receive what you declare. Anything you did not ask for is absent from your delegation, not present-and-empty.

🔴 fact_lookup_service works in-process and is withheld over the wire. It is a live service object, so it cannot be serialised to a remotely deployed agent — core drops it from the envelope unconditionally. The example above runs green under the harness and returns nothing once your agent is deployed remotely, with no error.

This is stated here because it is the one place where following our own documentation produces a silent failure. Facts are meant to reach a remote agent pre-loaded onto the envelope, declared in the agent's contract — that path is seamed but not yet wired, so a remote agent currently receives no pre-loaded facts. If you are building for remote deployment, do not design around fact_lookup_service yet; talk to us.

Two ways to write an agent, one runtime

@agent.handler is not a simplified mode you outgrow. A handler is compiled into a one-node graph, so it runs through exactly the same machinery as an agent you author as a graph yourself:

agent = Agent(name="…", description="…", graph=my_compiled_graph)

Start with a handler. Move to graph= when you need more than one node. You are not choosing a path you have to undo later.

Telling the model how to behave

🔴 The planner loop is yours, by design. This package ships nodes and conventions that drop into a graph you own. It does not ship a plan/act cycle, and that is a settled decision rather than a gap waiting to be filled — the topology, the edges and the iteration cap are things only you can size for your domain. So instructions=, routing=, narration_style=, tool_guidance=, prompt= and tools= are declarations: nothing feeds them to a model on your behalf.

agent.compose_prompt() is the method that turns them into a prompt. You call it and you get the assembled system prompt back, with your instructions verbatim inside it.

It takes no arguments, and there is no location field in the prompt — the prompt carries what your agent declares, and the turn's context comes from the turn's own inputs, which do not include a location.

One bundled example takes a different route, and you can check it by opening it: examples/train_search_agent.py calls d.generate_structured(...) inside a LangGraph cycle, with a prompt it writes itself. Whatever route you take, every model call goes through d.generate / d.generate_structured — see Calling a model.

Everything in this section describes what compose_prompt() returns.

Most agents need one thing — instructions=:

agent = Agent(
    name="weather_agent",
    description="Answers weather and forecast questions.",
    instructions="""
    You answer questions about weather and forecasts.
    Prefer official meteorological sources over aggregators.
    Never speculate beyond seven days — say you don't know instead.
    """,
)

Your instructions reach the model verbatim. We never rewrite them, and they are not a template: braces inside them are literal text, so you cannot accidentally interpolate one of our internal variables.

The SDK composes the rest of the planner prompt around them — the planning loop, how to ask a question and resume, and how to narrate.

The turn's own context travels separately: d.generate builds the user's language, locale, timezone and current time into a system message and prepends it to every call. That context is why this wrapper exists. Omit the language instruction and you answer a German user in English. Omit the locale rules and dates come out American. Omit the temporal instruction and "tomorrow" is the wrong day. None of these fail a test you would think to write, and all of them reach a user.

If you want to replace one composed section rather than all of it, pass narration_style=, tool_guidance= or routing=. If you want full control, pass prompt= — and swisper_agent_sdk.prompts.scaffolding() is importable, so you can still assemble the parts you did not want to write yourself.

📖 PROMPTING.md has the full map of what gets composed, plus two rules that are not general prompting advice — they are things we got wrong in production, in ways that reached users. Read it before you write your second agent.

It ships inside the installed package (next to examples/), so it is on your disk the moment pip install swisper-agent-sdk finishes. Find it with:

$ ./.venv/bin/python -c "import swisper_agent_sdk, pathlib; print(pathlib.Path(swisper_agent_sdk.__file__).parent / 'PROMPTING.md')"

Working from a source checkout instead? It is at src/swisper_agent_sdk/PROMPTING.md.

Using what Swisper already knows — d.recall

Your agent can ask for facts Swisper already holds about this user, so it does not have to interrogate them for things they have said before:

@agent.handler
async def handle(d: Delegation) -> Reply:
    facts = await d.recall("dietary restrictions")
    return Reply(text=f"I found {len(facts)} thing(s) I already knew.")

You declare the capability, and core decides what you actually receive:

agent = Agent(name="...", description="...", capabilities=("recall",))

Two outcomes that look similar and are not. Getting this wrong is the most likely way to ship a bug here, so the SDK keeps them apart rather than collapsing both into an empty result:

What happened	What you get
You did not declare recall	NotEntitledError — raised locally, offline, before any network call. Your agent never holds the delegation token, so the SDK mirrors the refusal core would give you anyway.
You declared it, and core found nothing you may see	An empty list. Not an error.

🔴 An empty list never means "you were refused." It means core looked and there is nothing here that this agent is allowed to see — which is a normal answer, not a failure, and your agent should carry on rather than escalate.

Offline, pass your own transport and no network is involved:

async def fake_recall(query: str) -> list:
    return [{"type": "dietary_restriction", "value": "vegetarian"}]

agent = Agent(..., capabilities=("recall",), recall_transport=fake_recall)

Calling d.recall(...) on an agent built without a recall_transport raises a RuntimeError that says exactly that — rather than silently returning nothing, which would be indistinguishable from "core knows nothing about this user."

Declaring a third-party credential your agent needs

If your agent needs a per-user connection to a third-party service, declare the credential kind it needs, as a plain string:

agent = Agent(name="…", description="…", requires_credentials=("example-audio",))

You declare only the key. Never a credential, never a scope, never a user — where and how a person logs in, and which hosts a token from that provider is allowed to reach, are trust decisions that stay with the operator. This is the same shape as capabilities= and cards=: your declaration only ever narrows what your agent is given, never widens it.

If you author your agent through agent.contract.yaml instead, the same list is a top-level credentials: section:

# agent.contract.yaml
credentials:
  - example-audio

The key must match a provider entry an operator has already added. A provider entry is core-owned, not something a partner registers by declaring it — you propose a new one to us, an operator adds it, exactly as agent activation already works. Declare a key with no matching entry and registration refuses your agent outright, with a 422 naming what was wrong:

{
  "code": "CREDENTIAL_KEY_UNKNOWN",
  "unknown": ["spotify"],
  "known": ["example-audio"],
  "message": "the contract declares a credential key core has no provider entry for; a provider entry is core-owned and operator-added -- a partner proposes it by declaring the key, an operator adds it"
}

unknown lists every key you declared that core has no entry for, and known lists every key that already has one — so the fix is either a typo in your own contract, or a new provider that needs to be proposed and added before you can declare it.

Calling a model

A submitted agent makes every model call through d.generate or d.generate_structured: Swisper's models, tiers and consumption policy. That is a security decision, made by design — it keeps inference in Switzerland or Europe and keeps your users' data from leaving. Calling a model directly from your agent, with a client of its own, is not supported. A model of your own is for testing on your machine only; your submitted agent never uses it and never holds a model key. swisper-agent check, and the developer page's Checks list, fail an agent that depends on or imports a direct model client — see A direct model client fails the check.

The platform holds the keys, picks the provider and meters the spend. You call one of two methods and never touch any of that:

from swisper_agent_sdk import Message

result = await d.generate([Message(role="user", content=briefing)], tier="planner")
print(result.text)

For an answer you can branch on, hand it a pydantic model instead and get a real instance back:

from pydantic import BaseModel

from swisper_agent_sdk import Message


class NextStep(BaseModel):
    tool: str | None = None
    answer: str | None = None


result = await d.generate_structured(
    [Message(role="user", content=briefing)], NextStep, tier="planner"
)

Both return a GenerateResult. d.generate fills .text; d.generate_structured fills .structured. Three more fields are on every result, whichever you called:

• result.model — the model that actually answered. When a run behaves oddly this is your only handle on why, and you cannot get it any other way. Log it.
• result.usage — this call's input_tokens and output_tokens. It is how the platform attributes cost to your agent, so it is never absent.
• result.capability_version — which version of the capability contract the platform answered with. You will not normally read it; the SDK checks it for you and raises CoreTooOldError if the platform is older than this SDK build needs, with the pin that would work today in the message.

🔴 .structured is a real instance of your model, or it is None. When the response carried a structured payload you get a validated NextStep — never a bare dict to parse yourself. When it did not, you get None. Handle both halves: reaching straight for result.structured.tool is an AttributeError waiting for one bad round, and nothing type-checks it for you.

You never name a language. A system message carrying the user's interaction language, locale, timezone and current time is composed and prepended to every generate call, automatically, both methods, every time. Do not build one, and do not translate anything by hand.

The three tiers

tier= is required and the vocabulary is closed to three names. Anything else raises InvalidTierError locally, before the call leaves your process, with the three legal names in the message — a typo costs you no round trip and no waiting.

tier	reach for it when
planner	The step reasons over the whole picture: deciding what to do next, weighing what a tool returned, writing the answer the user reads. The most capable, and the slowest.
tool_selection	The step is which of these, with what arguments — a bounded choice over a catalogue you supply.
fast_classification	The step is small and mechanical: routing an intent, picking between two obvious branches, tagging something. The cheapest and quickest.

🔴 It is tool_selection, never tool_calling. The tier picks a model that is good at deciding among tools. It is not a mode and not a second API: d.generate is the same call at every tier, and naming this tier does not by itself put a single tool in front of the model. Deciding which tool to call is your loop's job — show the model a catalogue in the prompt and read its choice off a typed result, the way examples/train_search_agent.py does. The old name read as "the tool-calling API", and a partner who goes looking for one loses an afternoon; the current name is the fix. If what you actually want is to hand the model tool declarations, that is a separate argument — see "Offering tools to the model", next.

Picking the right tier is a judgement no error can make for you. See PROMPTING.md, "Choosing a tier is a quality decision and nothing checks it" — it is the failure you meet after you know the vocabulary.

Offering tools to the model

🔴 tools= means two different things in this SDK, and they are unrelated. The tier vocabulary makes this collision easy to walk into, so here are both:

where	what it takes	what it does
Agent(tools=(my_function, …))	your Python callables	Prompt composition only. Their signatures and docstrings become the ## Your tools block in the prompt compose_prompt() builds. Nothing calls them.
d.generate(…, tools=[ToolDeclaration(…)])	ToolDeclaration objects	Goes on the wire. The declarations are sent with the request and reach the model as tools it may call.

A ToolDeclaration is a name, a description and a JSON-schema parameter dict:

from swisper_agent_sdk import ToolDeclaration

result = await d.generate(
    messages,
    tier="tool_selection",
    tools=[
        ToolDeclaration(
            name="search_connections",
            description="Find train connections between two stations.",
            parameters={
                "type": "object",
                "properties": {"origin": {"type": "string"}, "destination": {"type": "string"}},
            },
        )
    ],
)

No component in this package sends tool declarations for you yet, so the bundled example takes the other route: it shows the model a catalogue in the prompt and reads the choice off a typed result (examples/train_search_agent.py). Both routes work; the prompt-catalogue one is the one we have run in anger, and it is the one with a file you can copy.

Deadlines

Every call is bounded. Leave it alone and you get 30 s, sized for the slowest tier; pass deadline_ms= to set your own:

result = await d.generate(messages, tier="fast_classification", deadline_ms=5_000)

A call that does not come back in time is aborted and raises CapabilityTimeout, naming the tier, the deadline and the time actually spent. It is never left to hang — a slow provider on our side must not become an indefinite wait in your demo. If a tier legitimately needs longer, a bigger deadline_ms= is the answer.

Tightening the budget on a cheap tier is worth the one line: 30 s of a user waiting on a one-word classification is a defect no test will show you.

Watching the text arrive

d.generate() hands you the finished answer. If you want to show it as it is written, there are two ways, and they solve different problems.

The pieces, as they arrive:

async for chunk in d.generate_stream(messages, tier="planner"):
    print(chunk.text, end="", flush=True)
    if chunk.done:
        print(f"\n[{chunk.usage.output_tokens} tokens]")

The pieces and the finished answer, which is usually what you actually want:

result = await d.generate(messages, tier="planner", on_chunk=lambda c: show(c.text))
# result.text is the whole thing, already assembled

chunk.model rides every chunk; chunk.usage only arrives on the last one, because that is the first moment it is known. Nothing fakes an early value for it.

Streaming needs its own transport, separate from the one-shot one:

agent = Agent(..., stream_transport=my_stream_transport)

They are two seams because they are two shapes — one awaits a single response, the other yields frames — and a single parameter accepting both would let a client that cannot stream satisfy the type and fail at the first streamed call.

The deadline means something slightly different here

On a normal call, deadline_ms bounds the whole call. On a stream it bounds time without progress — the wait for the first chunk, and each gap between chunks.

It is the same rule, not a second one: a normal call has exactly one delivery, so "the whole call" and "the wait for the next piece" are the same number. Streaming is where they come apart, and bounding the total would mean a long answer that is arriving perfectly well gets killed for taking a while. What the deadline catches is a provider that has stopped, which is the failure you actually need protection from.

So a forty-five second answer that streams steadily is fine on the default. A stream that goes quiet for thirty seconds raises CapabilityTimeout.

What a generate call raises

🔴 The one to expect first is NoCapabilityConfigError, and it is the likeliest thing to meet you on your first live call. The platform resolves a model per (your agent, tier); when there is no row for that pair and no platform default either, the call is refused rather than quietly falling back to something nobody chose. The message is composed by the platform, which knows your agent's identity, and it names what an operator has to do about it. A tier that works for one agent can be unconfigured for yours — nothing about it is a property of your code.

raised	when	your move
InvalidTierError	tier= is not one of the three names. Raised locally, before the call leaves your process.	Fix the typo.
CapabilityRefused	The base of every refusal the platform sends back, and what you get for one this SDK build has no name for yet. .error_code is the stable code, .params the structured context, str(err) the platform's own sentence. Every class below it also is one.	except CapabilityRefused catches all of them; branch on .error_code when you need to.
DelegationTokenRefused	Your delegation token was refused: TOKEN_EXPIRED, SCOPE_NOT_GRANTED, INVALID_SIGNATURE, AGENT_MISMATCH, TOKEN_KIND_MISMATCH.	An expired token is fixed by re-minting. A scope you were never granted is not yours to fix — .params carries requested_scope and granted_scopes, which is the sentence that says which.
NoCapabilityConfigError	No model resolves for (your agent, that tier): no row and no default, no provider linked, the linked provider unknown, or the provider not serving that tier (.error_code says which of the four).	Send us the message verbatim — it names the row that is missing.
ProviderCallFailed	The provider call behind the capability failed on our side: an outage, a quota or rate window, a vendor timeout.	Wait and retry; .params["reset_at"] names when if the provider said. Quote .params["correlation_id"] if it persists.
CapabilityTimeout	The call outran its deadline and was aborted — locally, or the platform said DEADLINE_EXCEEDED.	Raise deadline_ms=, or move the step to a cheaper tier.
CoreTooOldError	The platform is older than this SDK build requires.	The message names the pin that works today.
ValueError	You passed stream=True to d.generate(). That method returns one assembled result; its transport awaits a single response. Raised locally, before any call is made.	Use d.generate_stream(...) for an async iterator of chunks, or d.generate(..., on_chunk=...) to watch text arrive and still get the assembled result.
CapabilityNotAnswered	The address answered, but not as Swisper: a web page, a proxy, or a backend that does not serve the capability path. str(err) names the URL, the status and content type that came back, and the settings to check; .url, .status and .content_type carry the same. The prompts capability raises it too, and the bundled-prompts fallback records it as the reason.	Point SWISPER_BASE_URL at the Swisper environment you submit to, or ask its operator to route /internal/v1/capability/ to Swisper.
RuntimeError	No generate_transport= was given to the Agent — or, served with create_app, neither SWISPER_CORE_URL nor SWISPER_BASE_URL is set.	See below.

Catching only the first and the timeout is the mistake worth naming: except (InvalidTierError, CapabilityTimeout) leaves the likeliest failure uncaught. All of these import from the package root, so you can except any of them by name. The platform's refusals are one family: a CapabilityRefused you did not anticipate is still a refusal with its code and params intact, never a KeyError from inside this package — and INFERENCE_MISCONFIGURED (a provider credential or configuration wrong on our side) arrives that way, with a message telling you not to retry.

⚠️ Until 2026-09-13 this table was wrong, and 0.1.0a17 — the last release before it was fixed — still behaves the old way: only NO_CAPABILITY_CONFIG was mapped, and every other refusal — an expired token, a scope you were never granted, a provider outage — came back as KeyError: 'model' raised from inside this package, indistinguishable from one another. If that is what you see, you are pinned to 0.1.0a17 or earlier. 0.1.0a18 is the first release that behaves as the table says.

Where the call goes, and the one thing you wire

d.generate does all of the call but the last hop: it composes the request, prepends the context block, checks the tier and enforces the deadline. (d.generate_structured does one thing more — it derives the JSON schema from your model and validates the response back into a real instance of it.) Carrying the request to the platform is a transport handed to the agent:

agent = Agent(name="…", description="…", generate_transport=transport)

The SDK ships that client, so you never write one against our wire shape:

from swisper_agent_sdk import http_generate_transport

agent = Agent(
    name="…",
    description="…",
    generate_transport=http_generate_transport(
        "https://core.example.com",       # core's ROOT — not the endpoint path
        token=current_delegation_token,   # a str, or a callable returning one
    ),
)

base_url is core's root; the endpoint path belongs to the SDK, so appending it yourself earns a 404 with nothing to explain it. token is the delegation credential — pass a callable in anything long-running, because tokens are TTL-bounded and a captured one stops working silently. "Connecting it to core", below, has the rest.

An Agent built with no transport raises a RuntimeError naming the missing generate_transport= the first time a handler calls d.generate. Served with create_app, it is given one instead, at SWISPER_CORE_URL, else SWISPER_BASE_URL (Serving your agent); with neither set, the RuntimeError names those two variables.

🔴 Supplying your own client=? Set timeout=None on it. httpx.AsyncClient defaults to a 5-second timeout. This SDK's deadline is 30 s. Leave that default in place and a legitimate thirty-second planner call dies at five with httpx.ReadTimeout — never becoming the CapabilityTimeout this SDK promises, and looking for all the world like our platform being flaky. The transport the SDK builds for you already sets timeout=None; the trap is only reachable by passing your own client. One deadline, and it is d.generate(deadline_ms=...).

That same seam is what lets a real planner loop run offline: pass a function that returns the response shape and your entire graph runs with no model, no key and no network. examples/train_search_agent.py does exactly that, and it is the file to copy.

Your prompts, and where they live in the Swisper ecosystem

instructions="..." is the fastest way to a working agent and fine for a single-node one. Past that, prompts belong in files — Jinja2 templates declared in your contract:

# agent.contract.yaml
prompts:
  - name: train-search.planner
    file: prompts/planner.j2

load_agent("agent.contract.yaml", ...) reads those files into the agent's library for you — the bundled copy is the default, never nothing. If you construct through contract.build_agent(...) instead (it has no path to read from), hand it the library yourself:

import pathlib

from swisper_agent_sdk import Agent, PromptLibrary
from swisper_agent_sdk.contract import AgentContract

contract = AgentContract.from_yaml(pathlib.Path("agent.contract.yaml").read_text())
agent = contract.build_agent(
    prompt_library=PromptLibrary.from_contract(contract, "agent.contract.yaml"),
)

Reading the operator's edits — declare prompts as a capability

⚠️ Available since 0.1.0a18 (2026-09-13 UTC). http_prompts_transport is included in the current pinned release, swisper-agent-sdk==0.1.0a30. It does not exist in 0.1.0a17 or earlier; see the red note at the top of this page.

Once registered, your prompts are managed in Studio: an operator edits them there, and the edit reaches your running agent through core (D-27). That read is a capability, exactly like recall and generate — your contract declares it, the operator's activation grants it, and the SDK reads only when it was declared:

# agent.contract.yaml
capabilities:
  - recall
  - prompts        # ← without this line, core will never serve you a prompt

from swisper_agent_sdk import current_delegation_token, http_prompts_transport, load_agent

agent = load_agent(
    "agent.contract.yaml",
    name="train_search", description="…",
    prompts_transport=http_prompts_transport(
        "https://core.example.com",       # core's ROOT — not the endpoint path
        token=current_delegation_token,   # the per-turn credential, see below
    ),
)

Contract says	What d.prompt(...) renders
prompts not declared	Your bundled .j2 files, always. Core is never asked, and nothing is announced — there was nothing to degrade from.
prompts declared, transport wired	The operator's current text, fetched through core on every turn. If core cannot serve it, the bundled copy — and the fallback is announced on the trace with its reason (prompt_fetch_fell_back_to_bundled), never silent.
prompts declared, no transport (local dev)	Your bundled files, silently — nothing in the process can reach core.
prompts declared and transport wired, but the contract ships no prompts: block	Nothing — d.prompt(...) raises, naming how to supply a library. With no declared files there is no bundled copy to render or to fall back to, and core is never asked: the absent library short-circuits the read before the transport is consulted.

🔴 Declaring the files is not declaring the capability. A prompts: block tells core which files you ship; capabilities: [prompts] tells core you may read the operator's edits back. Until 2026-09-12 the second was undocumented, and an agent that declared only the first rendered its packaged copy for ever while operators edited in Studio.

🔴 And the converse — declaring the capability is not declaring the files. That is the last row above, and it is the noisier failure of the two: the capability and the transport on their own give you an agent that raises the first time it renders anything.

🔴 contract.build_agent(...), not Agent(contract.name, contract.description, ...). The second builds an agent that knows its name and nothing else — no inputs, no capabilities, no cards, no entitlement key, no prompts. It works, so nothing complains, right up until you run the shipping step below: AgentContract.from_agent(agent) then emits a contract with eight of twelve fields empty, over the top of the one you wrote, and swisper-agent submit sends that. build_agent carries the whole declaration through and takes the same keyword arguments.

In a test, where you want to vary one prompt without touching a file:

agent = Agent("train_search", "finds trains",
    prompt_library=PromptLibrary.from_mapping({"train-search.planner": "..."}))

Now render it:

system = d.prompt("train-search.planner", destination="Zurich")
result = await d.generate([Message(role="system", content=system)], tier="planner")

A template branches on what the turn knows, and each branch says what to do about it:

{% if destination %}
The traveller is going to {{ destination }}. Do not ask again — confirm it and move on.
{% else %}
No destination yet. Ask before searching; a search without one wastes the turn.
{% endif %}

user_input, chat_id and current_time are always available without being passed. interaction_language, user_locale, user_timezone and llm_reasoning_language are available when the turn carries them. A value you pass explicitly always wins.

🔴 Treat those four as optional, because core does. Core sends every one of these keys on every turn and sends null for anything it has not resolved — a user with no locale on file, a channel that carries no timezone. Your template will not see a null; it will see nothing, and a bare {{ user_locale }} then fails that customer's turn while passing every test you wrote.

{{ user_locale | default('en') }}              {# good #}
{% if user_locale is defined %}…{% endif %}    {# good #}
{% if user_locale %}…{% endif %}               {# NOT a guard — raises on an absent name #}

That last line is the trap worth remembering: under StrictUndefined a bare truthiness test raises exactly as a bare reference does, so the obvious way to "handle the missing case" fails precisely when the case occurs. swisper-agent check recognises the two good forms and will tell you about the third.

current_time is the exception, and it is a rule rather than a special case: the SDK defaults what it can know for itself, and never invents what belongs to the user. Your process owns a clock. It does not own the customer's locale, and guessing one puts silently wrong formatting in front of a real person.

A typo fails the turn rather than rendering blank. Jinja's own behaviour for an unknown variable is the empty string, silently — one transposed letter would remove a section of your prompt and nothing would say so. Instead:

PromptVariableError: train-search.planner: destinaton is not available.
  available: current_time, destination, interaction_language, travel_date, user_input
  did you mean: destination

Checking every template before you ship

$ swisper-agent check --contract agent.contract.yaml
✗ travel_date — 'train-search.planner' reads it, this call site does not pass it

Offline, tokenless, non-zero exit — so it belongs in a pre-commit hook and in CI. You declare nothing: the allowlist comes from your own d.prompt("name", ...) calls. It also catches the reverse (a keyword no template reads) and templates nothing renders. The render-time error covers the branches your tests walk; this covers the top-level names in the ones they do not.

Read that boundary exactly, because it is narrower than it sounds. check verifies top-level variable names. It does not resolve attribute paths — claim.status, e.when — because those are properties of a runtime object and cannot be known from source. In a real template most references are attribute paths, so most of what your template reads is not statically checked. check prints every path it could not verify; read that list, and drive those branches in a test.

A template that will not parse is reported like anything else — naming the file, the line, and what Jinja was looking for — never as a stack trace:

✗ prompts/answer.j2:1 — Unexpected end of template. Jinja was looking for the
  following tags: 'elif' or 'else' or 'endif'.

check reads every .py beside your contract to find those call sites. If your layout has directories that should not count — vendored code, a generated client, fixtures that are wrong on purpose — name them:

$ swisper-agent check --contract agent.contract.yaml --exclude vendor --exclude fixtures

--exclude matches a whole path segment, so --exclude vendor leaves vendored_logic.py alone.

A .py file it reads that Python can't read fails the check with two lines of its own. The text after the line number is Python's own message:

✗ Python can't read broken.py:2: invalid syntax
      Fix the Python syntax error at broken.py:2, then run the checks again.

Code nested too deeply for Python's parser (a very long chain of +, say) gets two lines too:

✗ Python can't read deep.py: too deeply nested to check
      Split the code in deep.py into smaller expressions, then run the checks again.

check reads a .py file the way Python does: UTF-8, with or without a byte-order mark, or the encoding its # -*- coding: … -*- line declares. A file that is none of these gets one line:

✗ Python can't read latin.py: not UTF-8 text

A coding line that names an encoding Python does not know gets Python's own words in its place (unknown encoding: …).

Each counts as a problem, so check exits 2. The file is left out and every other file is still checked. The check for a direct model client (below) skips a file it can't read, so it cannot see that file's imports: fix that file first, then run check again.

A direct model client fails the check

check also fails an agent that calls a model directly rather than through d.generate / d.generate_structured (see Calling a model). It fails when the requirements.txt beside your contract names a direct model client, or when a .py file in that folder or below imports one — agent.py and every local module, an import inside a function included. The developer page's Checks list shows the same check, Your agent calls models only through Swisper, in the same words:

$ swisper-agent check --contract agent.contract.yaml
✗ Your agent calls a model directly: requirements.txt:3 names openai; helpers.py:1 imports google.genai (google-genai). A submitted agent calls models only through d.generate / d.generate_structured: Swisper's models.
      Use d.generate / d.generate_structured instead. Your own model is for local testing only (Try it, or swisper-agent dev --answers model).

Every finding is listed, each with its file and line: requirements.txt:<line> names <package> for a dependency, <file>:<line> imports <name> for an import, with the package in brackets when the import name differs from it. It counts as a problem, so check exits 2. The clients it knows:

openai, anthropic, google-generativeai, google-genai, mistralai, cohere, groq, together, litellm, ollama, langchain-openai, langchain-anthropic, langchain-google-genai, vertexai.

A name in requirements.txt matches however it is spelled or pinned (OpenAI==1.2, langchain_openai[extra]>=0.1), and on a line that points at a URL or a file it matches the name that line names: openai @ https://…, a VCS or URL line carrying #egg=openai (with -e or without), or a wheel by path or URL (langchain_openai-0.1.0-py3-none-any.whl). An import matches in any of its forms (import vertexai.generative_models, from google import genai). A comment or a string that mentions a client is not an import. --exclude does not narrow this check.

To fix it, make the call with d.generate / d.generate_structured and remove the client from requirements.txt. A model of your own stays on your machine, for local testing only.

The check is a guard, not the rule. It reads requirements.txt (not pyproject.toml, not a file that a -r line points to, and not a source archive named only by its path or URL) and import statements (not importlib or other imports made at run time), and it does not look inside environment, vendored or build folders (.venv, venv, node_modules, vendor, build, dist, __pycache__, hidden folders). The rule applies whatever the check sees: a submitted agent calls models only through d.generate / d.generate_structured.

Letting the person see the stream

on_chunk gives you the fragments. To let the user see them arrive:

result = await d.generate(messages, tier="planner", stream_to_user=True)

🔴 Each fragment is published one behind, and that costs one provider interval. The last text chunk has to be marked is_final, and nothing can know a chunk is last until the next frame arrives — so the publisher holds each fragment until it sees the following one. Measured with a 1-fragment-per-second stub: the model produced ONE at 1.0 s and the user saw it at 2.01 s. At a real model's 50–150 ms cadence that offset is invisible; for a provider that batches by sentence it is not. The alternative — publish immediately, mark nothing final — leaves every turn rendering as permanently still-typing, which is worse.

Note that result.published records content and order but not arrival time, so the assertions in this chapter stay green either way. If time-to-first-token matters to you, measure it on the wire.

One fragment at a time onto the channel Swisper already republishes to the frontend, with the last marked final. Without it, an agent can stream perfectly and the person still sees one block of text at the end.

Proving it, offline:

result = await harness.turn(agent, "trains to Zurich")
assert result.published                     # what the USER would have received
answer = [e.content for e in result.published if e.process_indicator is None]
assert answer == ["The ", "best ", "match…"]   # d.say events also land in `published`
assert result.published[-1].is_final

result.published is the only observable difference between an agent that streams to a person and one that streams into a void. Every other assertion — the text, the status, the timings — is green either way.

🔴 In a real deployment you must give the agent a sink your server can drain. The default is an in-memory collector, which is right for tests and wrong for a server: a list is only readable after the turn, so every fragment would arrive at once, at the end.

Serving your agent: swisper_agent_sdk.serve

Serve the agent with the SDK's own app. It does the streaming above for you, with a new StreamingEventSink for every request (🔴 ONE PER REQUEST, closed by the turn, never shared), and it does the part a hand-written handler forgets: it refuses every call Swisper did not sign, before your agent runs.

from swisper_agent_sdk import Agent
from swisper_agent_sdk.serve import create_app

agent = Agent(...)                 # no sink here: the app gives each request its own
app = create_app(agent)            # then: uvicorn my_agent:app

pip install uvicorn (or any ASGI server): the app is plain ASGI and pulls in no web framework. It serves POST /v1/execute and GET /health, and it already carries the tracing middleware, so do not add it again.

What it needs from its environment:

Variable	What it is
SWISPER_AGENT_PUBLIC_URL	Required. Exactly the URL you submit with submit --url, without /v1/execute: the address Swisper calls you at, and signs. It must be https://; http:// is accepted only for localhost, 127.0.0.1, ::1 and in-cluster service hosts (*.svc, *.svc.cluster.local). It may carry a path, but no query, fragment or credentials; anything else makes create_app raise ValueError, so the app does not start. Left unset, create_app still builds the app, but every call is refused NOT_CONFIGURED (the server.py that init writes refuses to start instead). It is never rebuilt from the request's Host.
SWISPER_BASE_URL	Where Swisper's public keys are, and where your agent's model calls go (d.generate / d.generate_structured), for an agent you run: the environment's address, e.g. https://dev.swisper.ai. It must be https://; http:// is accepted only for localhost, 127.0.0.1, ::1 and in-cluster service hosts (*.svc, *.svc.cluster.local). It may carry no query, fragment or credentials. Those keys decide which calls are Swisper's, so anything else makes create_app raise ValueError, and the app does not start.
SWISPER_CORE_URL	Not needed for an agent you run yourself: leave it unset and set SWISPER_BASE_URL. When it is set, the keys are read from it and the model calls go to it, instead of SWISPER_BASE_URL, under the same rule.
SWISPER_SUBMISSION_ID	Optional. When set, /health echoes it as submissionId; unset or blank, submissionId is null.

Your agent's model calls need no wiring here. An agent built without a generate_transport= is given the SDK's http_generate_transport at SWISPER_CORE_URL, else SWISPER_BASE_URL, with the token Swisper sends for each turn as its only credential; a generate_transport= you built the agent with is kept. With neither variable set, create_app logs it and every model call raises a RuntimeError: "this agent cannot call Swisper's models: neither $SWISPER_CORE_URL (set when Swisper hosts the agent) nor $SWISPER_BASE_URL (set it when you host the agent, e.g. https://dev.swisper.ai) is set". The host you give as SWISPER_BASE_URL must route /internal/v1/capability/* (and /internal/v1/providers/* for provider agents) to the Swisper backend; agents Swisper hosts reach it in-cluster.

Every /v1/execute must carry Swisper's RFC 9421 signature; a call without it, with an unknown key, older than 300 s, not matching its body or URL, or seen before is answered 401 with one of UNSIGNED, UNKNOWN_KEY, STALE, INVALID or REPLAYED, and your agent never runs. The module documentation (help(swisper_agent_sdk.serve)) has the full rules.

⚠ Writing your own /v1/execute instead means verifying that signature yourself, on every request. A handler that skips it answers anyone who can reach it.

Packaged UI results

An agent can return up to eight typed display outputs in its final reply:

return Reply(
    "Here is the overview.",
    ui_cards=[{
        "card_id": "overview",
        "schema_version": "1.0.0",
        "payload": {"label": "Current overview"},
        "fallback_text": "The interactive overview is unavailable.",
    }],
)

The card and exact schema version must exist in the reviewed package selected by core. swisper-agent card new makes that package (Adding a card). Payloads are finite JSON objects bounded to 256 KiB and nesting depth 32 (the root object is depth 0); fallback text is bounded to 2,048 characters. These fields carry display data, not user, release, resource or action authority. Core validates and captures them independently.

Use result.to_wire_result() only for the explicit final transport response. Ordinary model_dump(), model_dump_json() and representations omit the typed display channel. The SDK validates cards before the reply enters its traced graph, so payload and fallback do not become automatic trace output. This does not control logging independently implemented by an agent author. Existing d.card(...) markup and narration keep their existing behavior.

What tracing costs your turn

Publishing observations is synchronous on the request path, so an unreachable or wedged Studio backend is latency your customer feels. It is bounded — connection_timeout defaults to 1 second and is applied as both the connect and the read bound — but bounded is not free, and a turn publishes more than once.

Measured against a backend that accepts TCP and never answers, at the old 5-second default: the narration line arrived at +3,994 ms instead of +0 ms, and the result 4 seconds late. The answer itself was correct and complete throughout — a tracing failure never corrupts a turn — but the "still working on it" line exists to cover a wait, and it was late to its own job.

Raise connection_timeout if you would rather lose latency than lose traces. Do not raise it far without deciding that deliberately.

The two headers core sends you

Core passes the turn's trace context as X-Trace-Id and X-Parent-Observation-Id. TraceContextMiddleware reads them for you and you never handle them by hand — but they are what decides whether your agent emits observations at all, so it is worth knowing they exist. No header means no tracing, and that is the mechanism behind bring-your-own-key: core withholds them, and your agent emits nothing rather than inventing a trace of its own.

Two things in the snippet are load-bearing, and an earlier version of it got both wrong:

🔴 The turn closes the sink, in its own finally — not the reader. Iteration ends on the close sentinel and nothing else. A finally attached to the async for can only run after the loop, so it can never be what ends it: the request hangs forever, with no error and no log line. That snippet shipped, and a partner deploying a service hit exactly that.

🔴 One sink per request, and pass it to the turn — never bind it to a module-level agent. A shared sink gives every concurrent request the same queue, and whichever response generator reaches it first takes the fragment, including fragments belonging to someone else's conversation. Agent(event_sink=...) still exists for a single-tenant process; run_turn(event_sink=...) is what a server wants.

Add X-Accel-Buffering: no to the response headers if you sit behind nginx — otherwise it buffers your stream into one block and nothing in your tests will notice.

Seeing exactly what the model got

The prompt on the wire is not only what you wrote — Swisper composes the turn's language, locale, timezone and clock onto every call. That half is invisible in your own source, and it is usually the half you need when a model misbehaves:

result = await harness.turn(agent, "trains to Zurich tomorrow")
print(result.prompt)          # every message, role-labelled, exactly as sent

Why files, and not strings — the ecosystem answer

Because in a real Swisper environment, prompts are not a code artifact. They are managed content.

Swisper keeps prompts and configuration in a config service, and Studio is the surface people work through: it has a Template Editor that resolves a node's bound prompt, prompts are versioned, and publishing a new version re-points the binding to it server-side rather than asking every client to re-link. The same service holds the configuration around them — which model tier a node runs at, and the overrides that select it per agent and per node. You never set a model name in your code; you declare a tier, and the platform resolves what that means for your agent in that environment.

What that means for you, concretely, today:

• Write your prompts as .j2 files under prompts/ and declare them. swisper-agent submit packs the files themselves and sends them to Swisper. A prompt frozen in a Python string cannot make that journey.
• Locally, those same files are what render — no server, no credentials, nothing to run. What you debug against is byte-for-byte what submit sends, because it ships the files, not a copy of them.
• Once registered, a prompt is content that people other than you can see, version and edit — which is the point. The engineer is not the bottleneck on wording.

✅ That upgrade has landed — this paragraph used to say it had not. Until 2026-09-12 it read "editing a prompt in Studio does not yet reach a deployed partner agent without a redeploy", which contradicted "Reading the operator's edits" two sections above on the same page. The read path is built (D-27; the SDK side shipped in #2120): an agent that declares prompts in capabilities: and wires http_prompts_transport renders the operator's current text on every turn, no redeploy. That section is the authority on how; this one only explains why files were the right unit — declaring them as files is exactly what made the upgrade a configuration change for you rather than a rewrite.

What still holds: an agent that declares the files and not the capability renders its packaged copy for ever, and so does one with no transport wired. The capability is the switch, and it is not on by default.

Tracing — on by default, one line to wire up

Your agent traces itself. Every node, every model call, every tool call, published the same way an agent running inside Swisper publishes, so a turn of yours and a turn of ours look the same to whoever is debugging at 2am. There is no switch to turn it on, and that is deliberate: an agent nobody can see should not be something you can ship by forgetting a flag.

There is one line you do have to write, unless you serve with swisper_agent_sdk.serve, which already carries it.

from fastapi import FastAPI
from swisper_agent_sdk import TraceContextMiddleware

app = FastAPI()
app.add_middleware(TraceContextMiddleware)     # <- this one

Why that line is not optional, and what happens without it

Swisper has already opened a trace before it calls you, and it tells you where your work belongs in two request headers. Those headers arrive on the HTTP request and nowhere else — the delegation body does not carry them.

If those headers go unread, your agent does not appear in Studio at all. 🔴 Measured 2026-08-31, as a paired run: with the middleware absent the agent served the turn correctly and nothing whatever showed up — with it installed, the same turn appeared, nested under core's trace. Not a stray trace in the wrong place. Absent.

This paragraph used to say the opposite — that your agent "still traces, beautifully nested, under a root of its own, in a trace nobody is looking at". That was wrong, and wrong in the expensive direction: it sends you hunting for an orphan trace that does not exist. If you are debugging a missing agent, do not go looking for its trace elsewhere. Check this line first.

Either way nothing errors and no test you write will notice, which is the reason this is middleware you install rather than a function we ask you to remember to call.

It is plain ASGI, so it works under FastAPI, Starlette, Quart or a bare ASGI server, and it pulls none of them in.

Testing it, with nothing running

Configure nothing and your traces go into memory, where your tests can read them back. No Redis, no credentials, no network, no Swisper.

result = await harness.turn(agent, "book me a branch appointment")

plan = result.trace.node("plan")
assert plan.child("search_branches").type == "TOOL"    # nested, not a sibling

Assert on shape, not on counts. Tracing rarely breaks by stopping — it breaks by arriving flat: every observation present, every parent wrong, and the picture still renders. A count cannot see that. The second line above is the one that fails when it happens, and it fails nowhere else.

In production

Set SWISPER_STUDIO_REDIS_URL and SWISPER_STUDIO_PROJECT_ID in the environment and the same code publishes for real. Connection strings belong in your secret manager, so they are read from the environment rather than passed to a constructor — the argument exists if you need it, but you usually will not.

Two things worth knowing:

• Content is obfuscated before it leaves your process, always. Prompt and result bodies are hashed. There is no setting for this and we do not offer one, because the underlying default is off and an environment that simply never set it would publish customer content in the clear — a mistake with no symptom.
• A turn Swisper marks as private carries no trace header, and then nothing is published for that turn. Your agent answers exactly as usual. Absence of a trace means do not trace — never start one of your own.

If you build your graph yourself with Agent(graph=…) rather than @agent.handler, build it with create_traced_graph from swisper_studio_sdk instead of a bare StateGraph — otherwise the nodes in it are not traced. We do not reach into a graph we did not build.

Shipping your agent

agent.contract.yaml is your Agent(...) declaration as a file — identity, inputs, outputs, capabilities and cards — so core, or a teammate, can see what your agent promises without reading your source. It round-trips both ways:

from swisper_agent_sdk import AgentContract, load_agent

# emit it once your agent is built
AgentContract.from_agent(agent).write("agent.contract.yaml")

# read it back into a working agent — the file carries the declaration, never
# code, so you re-attach your graph/handler behaviour by hand
agent = load_agent(
    "agent.contract.yaml", name="my_agent", description="…", graph=my_graph
)

The declaration survives that trip intact — including prompts:, nodes: and edges:. (It did not in earlier alphas: from_agent returned those three empty, so from_agent(contract.build_agent()) silently discarded every managed prompt and your whole graph topology. If you wrote a workaround for that, you can drop it.)

No file at that path? load_agent(...) builds an agent exactly as it would without a contract at all — every DomainAgentInput field except fact_lookup_service reachable, unfiltered (default_inputs()), no declared outputs/capabilities/cards. Nothing regresses for an agent that has not adopted the file yet.

Submitting your agent — swisper-agent submit

Installing this package also installs a swisper-agent command. Getting an agent onto an environment is four commands, and nobody at Swisper has to do anything by hand until an operator reviews what you sent:

$ swisper-agent init --purpose "Finds train connections and says when they leave."
                                           # once: a runnable project (see below)
$ swisper-agent keys init                  # once per machine: your publisher key
$ swisper-agent keys contact --name "Ada Lovelace" --email ada@example.com \
    --organisation "Analytical Engines AG" # once: who Swisper may contact
$ swisper-agent submit --env dev --url https://my-agent.example.com

Nothing you submit is usable until an operator approves it in Studio. submit builds a package from your project, checks it, signs it with your key and sends it to the environment's door; the submission then waits for approval.

To show your agent's results as a card, add one before you submit: swisper-agent card new writes a first card and builds it, and swisper-agent card build rebuilds it after every change (Adding a card, below). submit sends it with the agent.

To have Foundry build your agent from its GitHub repository instead, press the developer page's Build with Foundry (Building with Foundry, below). The installer stores Foundry's address for you; if you installed the SDK another way, set it once with swisper-agent foundry set <url>.

swisper-agent register is retired. It needed a token nobody outside Swisper can get; it now prints where to go instead and exits 2.

swisper-agent init

Writes a runnable agent project to start from, ready for you to host (see What init writes below). It never overwrites, and refuses (exit 3, nothing written) when a path it would write — or a directory on the way to it — already exists or is a symlink.

• --purpose "TEXT" — one line saying what the agent does (at most 512 characters, no line breaks or control characters). It becomes identity.description in agent.yaml and the role line — the first line — of the agent's first prompt, prompts/plan.j2 (You are <name>. TEXT). Without it, both get a placeholder to replace. The prompt is declared in agent.yaml as <name>.plan, and agent.py renders it for its one model call. check refuses a declared prompt that no code renders, so remove the declaration if you remove that call.
• git. init then makes a git repository with a first commit of exactly the files it wrote — unless the directory is already inside a git repository (it says so and leaves that repository alone). The commit is made with your own git identity; when git config user.email is unset, the files are staged and init prints the git commit to run.
• GitHub. Then init creates a private GitHub repository named after the agent and pushes the first commit to it, with your own GitHub CLI (gh) and your own sign-in: Swisper never sees or holds a GitHub credential. Install gh from https://cli.github.com and sign in once with gh auth login. The repository belongs to the owner you chose last time, else your first organisation, else your own account; the owner a successful run used becomes the next default.
• --github-owner <login> — create the repository under this owner instead: your own login or one of your organisations. It must be a GitHub login (letters, digits and single -, not at either end, at most 39 characters); anything else is refused before anything is written (exit 2) and never reaches gh.
• --no-github — make no GitHub repository.
• --github — accepted and changes nothing, since the repository is made by default; it stays so that scripts which pass it keep working.
• --no-git — make no git repository. With nothing to push, there is no GitHub repository either, and init says so in one line.

A folder that already has its own git history. init only ever pushes the commit it made itself. If the folder is a git repository whose origin is on github.com, that is the agent's repository: init says so, and creates and pushes nothing. If it has no origin, an origin elsewhere, or the folder sits inside another repository, nothing goes to GitHub and init says why; commit the agent's files where that history lives, or create the agent in an empty folder.

When GitHub says no, the project is still there. If gh is not installed, you are not signed in, the owner already has a repository with that name, or the organisation does not let you create repositories, init says which, in plain words and with gh's own text where it gave one. Rate limits and other refusals show gh's own words. Either way the folder and its first commit are kept and init exits 0. Fix what it names, then run the line it prints, from inside the folder:

$ gh repo create <your-org>/my-agent --private --source . --push

init fills in the owner and the agent's name; it shows <your-org> when it could not yet tell the owner. When init could make no first commit (no user.email), make that commit first.

The agent init writes makes one model call. agent.py renders prompts/plan.j2 with d.prompt and asks the model with d.generate_structured (Calling a model), and its answer is the reply. It is the place to start changing: the prompt says what the agent is for, and the Answer model is the shape of what comes back. On the developer page, Try it asks you for that call's answer the first time (saved to answers.json), or your own model answers it when "Your model" is set. Once submitted, Swisper's models answer it. agent.yaml declares generate under capabilities: for it: Swisper grants a model call only to an agent that asks for it, so without it the call is refused once the agent is live, and swisper-agent check fails an agent whose code calls a model without it.

What init writes. Besides agent.yaml, agent.py, the first prompt, a skeleton ui/manifest.json, a README.md, a .gitignore and the two developer-page launchers, the project is ready for you to host as it stands:

• server.py — serves the agent with the SDK's own server, serve.create_app (Serving your agent), on port 8000: GET /health, and POST /v1/execute, which runs a turn only when Swisper signed the call. It binds the per-turn token Swisper sends (so the agent's capability calls work) and stops the turn when the caller disconnects. It refuses to start unless SWISPER_AGENT_PUBLIC_URL is an https:// URL; http://localhost and http://127.0.0.1 are accepted for local development only.
• Dockerfile and .dockerignore — an image that runs server.py with uvicorn on port 8000, as user 10001, with GET /health as its health check. .dockerignore keeps .env, key files (*.pem, *.key), .git, tests/ and answers.json out of the image, so pass SWISPER_AGENT_PUBLIC_URL with -e when you run it. The image also runs with a read-only file system (docker run --read-only --tmpfs /tmp …), as a hardened host would run it.
• requirements.txt — pins swisper-agent-sdk==<version>, the version that wrote the project, and adds uvicorn. Until that version is on PyPI, put its wheel (the swisper_agent_sdk-<version>-….whl file you installed it from) in a vendor/ folder in the project: the Dockerfile installs it from vendor/ when the folder holds it, and from PyPI otherwise. init does not create vendor/ for you.

Run it on your machine, from the project folder:

$ SWISPER_AGENT_PUBLIC_URL=http://localhost:8000 uvicorn server:app --port 8000

or, with Docker:

$ docker build -t my-agent .
$ docker run --rm -p 8000:8000 -e SWISPER_AGENT_PUBLIC_URL=http://localhost:8000 my-agent

Then http://localhost:8000/health answers {"status": "ok", "submissionId": null}. That is all a local server shows you: it refuses every call to /v1/execute that Swisper did not sign, and nothing on your machine signs one, so try the agent with swisper-agent dev or the developer page. Where you run it for Swisper, pass the address you submit instead — -e SWISPER_AGENT_PUBLIC_URL=https://<where this agent is reachable> — and set SWISPER_BASE_URL too (Serving your agent).

Trying it on your machine — swisper-agent dev

$ swisper-agent dev --agent agent:agent --answers answers.json

A local imitation of Swisper: type a message and press Enter to run a turn, and end with Ctrl-D (or pipe the lines in). A line typed while the agent is waiting on a question answers it. Only Swisper's side is simulated: every model and provider call is answered from the --answers file (JSON) and shown as simulated, and a call the file has no entry for fails the turn rather than being made up. Your agent's own network, database and file I/O run as written. Each call is one line. Only a call that failed for want of an entry adds (answers: <file>) after "the turn fails", naming the answers file it looked in ((answers: no answers file) when no --answers was given); a call that was answered, or that failed for another reason, has no such suffix.

--answers model. Your model (the one Try it's Your model mode uses) answers the model calls, shown as not simulated (<model>). It is used only on this machine, by Try it and dev, never by the submitted agent. Set it with swisper-agent model set (Your own model for local testing, below). Provider calls are still answered from a file: the answers.json in the folder of your agent's module, and only when that folder holds agent.yaml or agent.contract.yaml (never the current folder's). That folder is the project, and a provider call's line also shows what the developer page shows for it: the provider's display name, when the contract's providers: entry gives one (display_name), and whether the answer matches the operation's outputSchema in catalogue/operations.json, when the catalogue gives one:

  · simulated POST /internal/v1/providers/home-audio/operations/get_volume [<node>] — answered from the script (provider entry 1) — provider: Home Audio (home-audio) — schema: matches get_volume's outputSchema in catalogue/operations.json

An answer that breaks its schema ends — schema: does not match … instead; the answer is still given to your agent, and the mismatch alone does not fail the turn: a run whose turns complete still exits 0. [<node>] is shown only for a call made inside a graph node.

When the project's answers.json is not there, a provider call fails the turn and its line says so:

  · simulated POST /internal/v1/providers/<key>/operations/<op> [<node>] — no scripted answer — the turn fails (answers: <folder>/answers.json, which is not there)

and the run's first line ends "provider calls answered from /answers.json, which is not there." When the folder holds no contract, it is not a project: the file is not read, and the run's first line says why, once: "provider calls answered from /answers.json, not read: holds no agent.yaml or agent.contract.yaml, so it is not a project." Each provider call then fails the turn, and its line ends — no scripted answer — the turn fails (answers: none — not a project). A file actually named model is passed as ./model.

It exits 0 when every turn completed or is waiting on a question, and 3 when any turn failed (nothing was sent). It exits 2 before any turn runs for a missing or unusable answers file (not there, not JSON, or an entry it cannot read), whether given with --answers or, in --answers model, the project's answers.json; and for bad usage, such as an --agent that does not load. --answers model with no model set says how to set one and exits 3, as it does when your model settings cannot be read.

Your own model for local testing — swisper-agent model

$ swisper-agent model set --preset gemini --model <model id> --key   # the key at a hidden prompt
$ swisper-agent model show                                           # what is stored, never the key
$ swisper-agent model test                                           # two tiny calls, to see it answer

These settings name the model that answers your agent's model calls while you try it on this machine: in Try it on the developer page, whose Your model mode uses them (its settings page, Your own model, edits the same settings), and in swisper-agent dev --answers model. That is all they are for.

🔴 Used only on this machine, only for testing. The submitted agent never uses these settings: the SDK's server (swisper_agent_sdk.serve) never reads them, and the agent never holds a model key. In Swisper, every model call your agent makes goes through d.generate or d.generate_structured (Calling a model).

Where they live. In the SDK's config directory, the one that holds your publisher key (swisper-agent keys), mode 0700: model.json holds the style, preset, address, model and per-tier models, never the key; model-key holds the key alone. Both are mode 0600. They are never in the project, never in answers.json, and never sent to Swisper. model show prints the directory.

model set changes the stored settings. A flag not given keeps its stored value; with nothing stored yet, it starts from the default preset, gemini.

• --preset ID fills the style and the address: openai, gemini (the default), ollama, lmstudio or anthropic; none for an address you type. ollama and lmstudio run on this computer and take no key. Presets name no models: ask the provider for its current ones with model test --list.
• --style openai|anthropic: openai for OpenAI-compatible Chat Completions, anthropic for the Messages API.
• --address URL: the provider's API address, https://…, or http:// only for a model on this computer (localhost, 127.0.0.1 or [::1]). No user name or password, no query (?) and no fragment (#): the key goes in the key, never in the address.
• --model ID: the model that answers every tier. --tier TIER=MODEL gives one tier (planner, tool_selection or fast_classification) a model of its own, and TIER= hands it back to the main model. It can be repeated.
• --key asks for the key at a hidden prompt; --key-stdin reads it from stdin, for a script or a password manager; --clear-key deletes it. Use one of the three at most. The key is never taken from the command line: a value after --key is refused without being repeated, and nothing is saved.
• The key belongs to the address it was typed for. Changing the address needs a new key (--key or --key-stdin) or --clear-key; otherwise the save is refused on the key.

It prints ✓ model settings saved, then the settings as show does. Each refused setting is one ✗ <field>: <why> line, followed by Nothing was saved.

model show prints the style, the preset, the address, the model, each tier's model, key stored: yes or no, and where the settings are stored. The key is never printed. With nothing stored, it says so, how to set them, and the presets.

model test calls your model with the stored settings and key: one short text call, then one tiny structured call, the way Try it calls it. --model ID tests another model without saving it. --list asks the provider for its models instead, which checks the key and the address. It prints ✓ ok — … when the model answered (or the provider listed its models), else ✗ <kind> — <why>, where the kind is one of:

• key: the provider refused the key (HTTP 401 or 403);
• structured: the model answered text, but not in the structured shape Try it asks for;
• model_error: the provider answered with an error, shown as it gave it;
• unreachable: nothing answered at the address, or it is not an API there;
• rate_limited: the provider is rate-limiting the key (HTTP 429);
• timed_out: no answer within 10 s.

Not supported: your model must speak OpenAI-compatible Chat Completions or Anthropic's Messages API, with an API key. Providers that need their own cloud sign-in (AWS Bedrock, Google Vertex, Azure OpenAI's own auth) are not supported.

model exits 0 when it showed or saved the settings, or the model answered; 2 for a usage error, a setting set refuses, a key given on the command line, or no key at the prompt or on stdin (nothing is saved); 3 when the settings cannot be read or written, or test finds none stored; 4 when the provider refused (the key, the structured shape, or its own error); and 5 when no answer came (unreachable, rate-limited or timed out).

Adding a card — swisper-agent card new and swisper-agent card build

A card is what Swisper shows in the chat for a result of your agent: one self-contained HTML file that draws the data it is handed. Run both commands in the agent's folder (the one with agent.yaml). They work offline, with no token and no network: nothing is installed and nothing is fetched (no npm). The build is a zip of your own files.

$ swisper-agent card new                   # a first card, named after the agent
$ swisper-agent card new --name my-card    # or with a name you choose
$ swisper-agent card build                 # rebuild after every change to a card

What card new writes, then prints, with the next steps:

• ui/src/<card>/card.html — the card. It loads nothing from anywhere (no <script src>, no <link href>), and Swisper refuses a card that tries. It talks to the page that hosts it in JSON-RPC messages (MCP Apps), and the handshake is already in it: the card asks ui/initialize, the host answers, the card says ui/notifications/initialized. The host sends the data only after this handshake, and closes a card that skips it after a few seconds. The card shows the data as text, never read as HTML, whatever it holds.
• ui/src/<card>/fixtures/ready.json — the example data the developer page previews the card with.
• ui/manifest.json — declares the card: its name, its address ui://<agent>/<card>, its file, the data it takes (payloadSchema, a title and a text to start with) and its example data. An existing manifest keeps its versions and its other cards as they are; a project without one gets one, with the contract's version.
• ui/build.py — when the project has none (see below).
• ui/package.zip — the built card package. Its digest goes into the manifest as packageDigest.

The flags:

• --name <card> — the card's name: a lowercase letter, then lowercase letters, digits, - and _, at most 64 characters. Without it, the card is named after the agent. connect (or swi-connect) is the name of Swisper's own card and cannot be yours: an agent named connect gets connect-card by default. When the agent's name cannot be a card name (it starts with a digit, for example), the default is card.
• --contract <path> — the agent's contract (default ./agent.yaml). card build takes it too.

An agent has at most 16 cards; card new refuses a seventeenth.

Refusals write nothing. card new never overwrites. It lists what is in the way and writes nothing at all when a file it would write is already there, the manifest already has a card with that name or already holds 16, ui/build.py is a build of your own (add the card there), ui/package.zip was built some other way, the manifest cannot be read or names card files that are not under ui/src/, or something on the way is a symlink. Give the card another name with --name, or move those files first.

card build is the rebuild step. It zips every file under ui/src/ into ui/package.zip and writes the package's digest into ui/manifest.json, changing nothing else there. The package is reproducible: the same files always give the same packageDigest. Hidden files (a name starting with ., such as .DS_Store) are left out, and a symlink is refused. When ui/src/ holds no package.json or lockfile, the package gets a package.json and an empty lockfile, which say the cards have no npm dependencies. It refuses, writing nothing, when there is no ui/manifest.json, or a card in it names a file or example data that is not under ui/src/.

python ui/build.py does the same, as a convenience, with a Python that has the SDK. With one that does not, it says to run swisper-agent card build instead.

The loop. The developer page's Card screen (swisper-agent ui) shows every card in your package, in every state its example data gives, read from ui/package.zip — so a change shows only after a build:

1. edit card.html, its example data and, when the data changes shape, its payloadSchema in ui/manifest.json;
2. run swisper-agent card build;
3. reload the developer page.

card exits 0 when it wrote or built the card, 2 when there is no agent.yaml there — and card new also when it cannot be read, the agent's name cannot be part of a card's address, or --name is not a card name or is Swisper's own — and 3 for any refusal above. Sending the card from your agent is Packaged UI results; submit sends the package with the agent.

Building with Foundry — swisper-agent foundry

Foundry can build and submit your agent for you, from its GitHub repository. The developer page (swisper-agent ui) hands the agent over with one button, Build with Foundry; swisper-agent foundry tells that button where Foundry is.

$ swisper-agent foundry set https://foundry.example.com   # once: Foundry's address
$ swisper-agent foundry show                              # the address in use, and where it is from
$ swisper-agent foundry clear                             # delete the stored address

The installer sets Foundry's address for you. It stores https://foundry.swisper.ai when no address is stored, and keeps one you set yourself, also when you run it again. Change it with swisper-agent foundry set <url>, or override it with the environment variable SWISPER_FOUNDRY_URL. Without an address, the button stays disabled.

• set <url> checks the address and stores it, without a trailing /, replacing any address stored before. It accepts https:// with a host, or http:// only for this computer (localhost, 127.0.0.1 or ::1); no user name or password, no query (?), no fragment (#) and no spaces. The address is opened in your browser, so nothing that could hide a second destination or a secret is accepted. A refused address prints why, then Nothing was written. The address is stored as foundry_url in the SDK's config directory (the one that holds your publisher key, swisper-agent keys), mode 0600, readable by you alone, and never in the project.
• show prints the address in use and where it comes from, as below. With neither a variable nor a stored address, it prints foundry: not set and how to set one.
• clear deletes the stored address. Clearing when nothing is stored is not an error.

SWISPER_FOUNDRY_URL wins over the stored address, under the same rule, and show says which one is in use:

$ swisper-agent foundry show
foundry: https://foundry.example.com
from:    config (<config directory>/foundry_url)
$ SWISPER_FOUNDRY_URL=https://foundry.example.com swisper-agent foundry show
foundry: https://foundry.example.com
from:    SWISPER_FOUNDRY_URL (env; wins over a stored address)

set and clear still change the stored address, and say so when the variable is set and wins. An address that breaks the rule counts as none. That holds for the variable, and for a stored file that others can read or that no longer holds an address. A variable that breaks the rule is not passed over in favour of the stored address: fix it or unset it. The page then treats Foundry's address as not set, so it never offers a dead or unsafe link, and show says what is wrong.

foundry exits 0 when it stored, showed or cleared the address, including show with nothing set and clear with nothing to clear. It exits 2 for an address set refuses, or a missing argument. It exits 3 when show finds an address it cannot use, or when the config directory cannot be read or written.

The page's Build with Foundry button. It opens a link to Foundry. Foundry is never called, and nothing comes back. The link is built on this computer from three things only:

• the folder's GitHub repository, <owner>/<name>, from its git origin;
• the Swisper installation the agent is submitted to. It is derived from the same address the page's Submit sends to (--core-url when given, else the dev environment's https://dev.swisper.ai), reduced to scheme, host and port, in lower case. A user name, password or token inside --core-url is never shown on the page and never put in the link, and swisper-agent submit and status never print one in the terminal either. If that address cannot be reduced this way, there is no installation to name, and pressing the button is refused (below);
• this SDK's version.

https://foundry.example.com/handover/swisper-agent?v=1&repo=<owner>/<name>&installation=https%3A%2F%2Fdev.swisper.ai&sdk=<version>

The link carries no key, token, session or contact details. Foundry reads everything else, such as the agent's name and purpose, from the repository. It signs and submits with its own organisation key, not yours.

When you press the button, the page records which folder was handed over, and when (in UTC). The record is foundry-handovers.json in the SDK's config directory (mode 0600), keyed by the folder's absolute path, and never goes in the repository. So the page still knows after a reload or a restart, and pressing the button again updates the time.

When the button is disabled, the page shows why, and nothing is recorded:

• No GitHub repository. The folder's own git origin is none of https://github.com/<owner>/<name>[.git], ssh://git@github.com/<owner>/<name>[.git] or git@github.com:<owner>/<name>[.git], or there is no git repository or no origin. The host may be written in any case; <owner> must be a GitHub login (letters, digits and single -, not at either end, at most 39 characters). Only the folder's own repository counts: a folder inside someone else's checkout is not that checkout's repository. init creates the repository by default (swisper-agent init). A user name or token in an https origin is dropped before anything reads it, so it never reaches the link.
• No Foundry address. The page shows how to set one: run swisper-agent foundry set <url>, or set SWISPER_FOUNDRY_URL.

When both apply, the page names the missing repository. A folder with no agent in it has nothing to hand over.

When pressing the button is refused. Two more problems are found only when the button is pressed. The page's hand-over request then answers 409 with the code below and a message saying why. No link is returned and nothing is recorded. Both are checked after the two above, which win (FOUNDRY_NO_REPO, then FOUNDRY_NOT_SET):

• FOUNDRY_BAD_CORE_URL: the Core URL is unusable. The address Submit sends to is not an http:// or https:// address with a host and a valid port, so it cannot be reduced to scheme, host and port. The message never repeats the address. Start swisper-agent ui again with a correct --core-url.
• FOUNDRY_RECORD_FAILED: the hand-over cannot be recorded. foundry-handovers.json cannot be written in the SDK's config directory, for example when SWISPER_AGENT_CONFIG_DIR is a symbolic link.

swisper-agent keys — your publisher key

Your key is your identity as a publisher. Nobody at Swisper mints you a credential: Swisper learns your public key from your first signed request, and an operator approves what you submit. Every submit and status request is signed with it.

$ swisper-agent keys init                  # create it (refuses if one exists)
$ swisper-agent keys show                  # its fingerprint, and the predecessor's
$ swisper-agent keys export --to ~/safe/swisper-publisher.pem   # back it up
$ swisper-agent keys rotate                # replace it; the old key vouches for the new
$ swisper-agent keys contact --name N --email E --organisation O   # your details
$ swisper-agent keys contact               # show them

The key lives in $SWISPER_AGENT_CONFIG_DIR, else %APPDATA%\swisper-agent on Windows, else $XDG_CONFIG_HOME/swisper-agent, else ~/.config/swisper-agent — as keys/publisher-ed25519.pem, mode 0600 in a 0700 directory. It is never printed. The fingerprint (lowercase hex SHA-256 of the raw public key) is what to compare when someone asks "is that you?".

Back it up with keys export. Lose the key and you lose the identity, and with it everything Swisper approved for it. The backup is written 0600 and never overwrites an existing file.

Rotating. keys rotate makes a new key, has the old one sign it, and keeps the old key under keys/archive/. Your next requests carry that vouch, so Swisper treats the new key as the same publisher — including a block, which a rotation never escapes.

🔴 Make one signed request (submit or status) with the new key before rotating again. Only the immediate predecessor vouches. A key rotated twice before Swisper saw the middle one arrives unvouched, and starts unapproved.

Your contact details. Every submission carries your name, email and organisation, so the operator reviewing it knows whom to ask. Set them once with keys contact; all three are required. The email is checked for its shape only (you@example.com); name and organisation are 1–256 characters on one line. A refused value names the field and writes nothing (exit 2). They are stored beside the key as contact.json, mode 0600, and a rotation keeps them.

swisper-agent check --url — before you submit an address

Once your agent runs somewhere public, ask whether Swisper will be able to call it there. Run it from the project folder:

$ swisper-agent check --url https://my-agent.example.com

The project is the current folder, whose contract swisper-agent ui would serve, or, with --contract path/to/agent.yaml given too, that contract's folder; the .env there is the one public_url below reads. --contract may also name the project folder itself, which must then hold agent.yaml or agent.contract.yaml; a folder with neither is refused before anything is called, exit 2: "✗ is a folder with no agent.yaml or agent.contract.yaml in it; pass its agent.yaml." One of --url and --contract is required: with neither, check exits 2 ("one of the arguments --contract --url is required").

It asks the address three questions, in this order, and prints one line for each — ✓ or ✗, the check's name, and what it found in plain words:

• https — the address answers over TLS with a certificate this computer trusts: the public authorities, or SSL_CERT_FILE / SSL_CERT_DIR when set (and then it says so, because that bundle is on this computer, not where Swisper calls from). A failure says which: a certificate this computer does not trust ("use a certificate from a public authority, such as Let's Encrypt", and, for an unknown issuer, that a network which replaces certificates on the way, such as a company proxy, causes this too), a connection that never completes a TLS handshake ("Is https served on this port?"), a name that does not resolve, or nothing answering.
• health — GET <address>/health answers 200 with JSON whose status is "ok", as an agent served with the SDK does. A failure names a redirect ("Swisper does not follow redirects: submit the address your agent answers at itself"), any status other than 200, an answer that is not JSON, or a status other than "ok".
• public_url — an unsigned POST <address>/v1/execute is refused 401 UNSIGNED, and this project's SWISPER_AGENT_PUBLIC_URL (the environment variable, else that key in the folder's .env) is the address. A failure names which: the agent answered NOT_CONFIGURED (it was started without SWISPER_AGENT_PUBLIC_URL, so it refuses every call Swisper makes: set it to the address and restart it), it answered anything other than 401 UNSIGNED, the project sets no SWISPER_AGENT_PUBLIC_URL, the project's value is not usable, or it differs from the address.

When https fails, nothing else is called and the other two read "not checked". health and public_url ask different routes, so each runs whatever the other found. A last line, marked ℹ, says what this computer cannot check: whether Swisper's network reaches the address, whether your agent reaches Swisper back (it fetches Swisper's signing keys from there), and whether the running agent's SWISPER_AGENT_PUBLIC_URL is this project's value, beyond the fact that it has one.

An address that is not https is a verdict, not an error. Nothing is called, and the answer looks like this (http://localhost included):

$ swisper-agent check --url http://my-agent.example.com
✗ https      Swisper calls an agent only at an https:// address, and this one is http://. Nothing was called.
✗ health     The address is not https, so this was not checked.
✗ public_url The address is not https, so this was not checked.

  ℹ This computer cannot check whether Swisper's network reaches the address, nor whether your agent reaches Swisper back (it fetches Swisper's signing keys from there). Nor can it see whether the running agent's SWISPER_AGENT_PUBLIC_URL is this project's value: at most that the agent has one.

Every call goes one at a time to the address, never to Swisper, and takes at most 10 s of wall-clock time from connecting to the last byte read. Redirects are not followed, the calls are never signed, and they never carry a token or the address's user:password@. It connects directly, never through a proxy set on this computer.

It exits 0 when all three checks pass and 3 when any fails, a non-https address included. An address it cannot read as a URL with a host, or one that names port 0, exits 2 before anything is called, as do a --contract folder with no contract in it and --url together with --state or --exclude (those belong to check --contract).

swisper-agent submit

$ swisper-agent submit --env dev --url https://my-agent.example.com

Coming soon: Swisper hosts your agent. Today you run your agent yourself and submit the address it answers at. Swisper running it for you, from a signed and scanned image, is coming.

• Where it runs. --url https://… for an agent you run — and serve it with SWISPER_AGENT_PUBLIC_URL set to exactly that URL, the one that goes into runtime.json — with Docker, docker run … -e SWISPER_AGENT_PUBLIC_URL=https://my-agent.example.com …. The server.py that init writes refuses to start without it (or with anything but an https URL; http://localhost and http://127.0.0.1 for local development only), and serves with the SDK's own swisper_agent_sdk.serve, which refuses every call Swisper did not sign for that URL (Serving your agent). Check the address with swisper-agent check --url before you submit it. submit warns when the variable is set here and differs.
• Where it goes. --env dev — the only environment today. --core-url URL points at a local stack instead.
• The project. --contract (default ./agent.yaml, else the older ./agent.contract.yaml); its directory is the project.
• Tiers. --tiers basic,pro names the tiers you ask for. The default is basic,pro. The operator may grant any subset of what you ask for, but granting more than you asked needs the Access console — so the default asks for both, and the operator narrows it.

What goes into the package, from the project directory:

Entry	From
agent.yaml	your contract, byte for byte, whatever its file name
prompts/…	every file under prompts/ (dot-files skipped; a symlink is refused). A prompt the contract declares outside prompts/ is refused.
catalogue/operations.json	if present. Every provider it has operations for must be declared under the contract's providers: with endpoint: — the provider's MCP relay URL, https, on one of that provider's own hosts. check and submit refuse the package otherwise.
ui/manifest.json + ui/package.zip	only when ui/package.zip exists. A manifest without it submits the agent card-less, with a notice — and its versions must still match the contract's. swisper-agent card new writes both; swisper-agent card build rebuilds the package and its digest in the manifest (Adding a card).
runtime.json, intended_tiers.json	the flags above
submitter.json	your contact details from keys contact: {"name", "email", "organisation"}
evidence/check.json, evidence/doctor.json	check run on your contract; doctor's verdict and exit code when you pass --agent module:attribute. A doctor that cannot load the agent refuses locally, and its error stays on your terminal.

A symlink anywhere submit reads from — the contract, prompts/, catalogue/, ui/, or any directory on the way to them — is refused before anything is read or sent (ENTRY_SYMLINK), so a cloned repository cannot make submit upload a file from elsewhere on your machine.

Without contact details submit refuses before building anything (exit 3) and says to run swisper-agent keys contact --name … --email … --organisation …. Swisper keeps the details a version was submitted with: if yours changed since you submitted this version, submit refuses before sending and says so — they travel with your next version, so bump identity.version.

Before anything is sent, submit runs check (a failing check refuses), builds the package and validates it with the same validator the door runs — so a package the door would refuse never leaves your machine. The package is deterministic: submitting an unchanged project again sends identical bytes, and Swisper answers with the same submission rather than a new one. Anything that changes a byte makes it a second submission of the same version, which Swisper refuses (VERSION_EXISTS) while the first one waits.

Validate a package only with swisper_agent_sdk.package.validate_package — the function submit and the door both call. It enforces size, depth, count and pattern-length limits that the shipped submission_schema.json alone does not, so validating a package view directly against the schema is unsupported.

A connection failure is retried up to three times (after 1 s, then 2 s). A 503 means Swisper is busy (for example CHECKS_UNAVAILABLE, when its checks are all running) and resending the same package is safe, so it is retried up to three times over about 20 s (after 5 s, then 15 s, or after the Retry-After Swisper sends, if 30 s or less), and submit says it is waiting. A 429 is retried once, only when Swisper says to wait 30 s or less. Every attempt is signed afresh. Nothing else is retried.

On success it prints the submission id, its state and the next command:

✓ submitted my_agent 1.0.0 to https://dev.swisper.ai
  id:      3f2c9a4e-8d1b-4c7e-9a2f-5b6d7e8f9a01
  state:   waiting for approval

Next: swisper-agent status 3f2c9a4e-8d1b-4c7e-9a2f-5b6d7e8f9a01

swisper-agent status

$ swisper-agent status <id>
$ swisper-agent status --agent my_agent    # every submission of it that your key sent

Prints the state (received, checks failed, waiting for approval, approving, approval stopped at step N, live, rejected, suspended, superseded), the step, the reasons, the operator's note or rejection reason exactly as written, where the agent runs, and when it last changed. Only the key that submitted can read a submission. --env defaults to dev.

Exit codes

The exit code tells you who said no, which matters because the fix is different in each case:

Code	Meaning	What to do
0	accepted (submit) / read (status) / every check passed (check, check --url) / the card written or built (card) / the address stored, shown or cleared (foundry) / every turn completed or waiting on a question (dev) / the model settings shown or saved, or the model answered (model)	continue
2	usage error — a bad or missing argument, such as an address check --url cannot read as a URL, a --name that cannot be a card's, an address foundry set refuses, a --contract folder with no contract in it for check --url, a missing or unusable answers file for dev (its --answers, or in --answers model the project's answers.json), or a setting model set refuses or a key given on its command line (and the retired register). For check (without --url), 2 also means a check failed: a usage error and a failed check exit alike, so a script treats both as a stop	fix the command, or what the check names
3	refused locally — nothing was sent (a failing check, an invalid package, no key, no contact details, or changed details for a version already submitted; for check --url, a check that failed; for card, a refusal, which writes nothing; for foundry, an address it cannot use, or a config directory it cannot read or write; for dev, a turn that failed, or --answers model with no model set or with model settings it cannot read; for model, settings it cannot read or write, or model test with none stored)	fix what it names
4	refused by Swisper — the request was sent and declined; the code is printed verbatim (e.g. PUBLISHER_BLOCKED, VERSION_EXISTS, RATE_LIMITED). For model test, your model's provider refused: the key, the structured shape, or its own error	read the reason; retrying changes nothing (for model test, fix the key or setting it names)
5	Swisper unreachable, or its answer could not be trusted — no answer, a 5xx after the retries, or a 2xx whose body does not prove the outcome (unreadable, or missing the submission's id/state). For model test, no answer from your model's provider: unreachable, rate-limited or timed out	for model test, check the address or try again later; otherwise run swisper-agent status --agent <name> first: if it is listed, it arrived. If not, submit again later

🔴 4 and 5 are deliberately different. A refusal is a decision and an outage is weather; a pipeline that treats them alike either retries refusals forever or fails a deploy over a blip.

🔴 fact_lookup_service can never be declared in agent.contract.yaml. It is a live service object, not data — core withholds it from the wire unconditionally, for every agent, with no per-agent override. Declaring it here raises WireUnserialisableInputDeclaredError, on both the write and the read path. See the fact_lookup_service note above: it works for an in-process agent (examples/meal_planning_agent.py), never for one declared to run remotely.

Connecting it to core — http_generate_transport

Any agent that calls a model needs a transport: the client that speaks to core's capability endpoint. The SDK ships one, so you never write an HTTP client against our wire shape. Served with create_app, an agent built without a transport is given it, with the per-turn token bound for you (Serving your agent); what follows is for wiring it yourself:

from swisper_agent_sdk import Agent, http_generate_transport

agent = Agent(
    name="train_search_agent",
    description="Finds train connections.",
    graph=my_graph,
    generate_transport=http_generate_transport(
        "https://core.example.com",   # core's root — not the endpoint path
        token=current_delegation_token,   # a str, or a callable returning one
    ),
)

Then await d.generate(messages, tier="planner") reaches a real model.

Pass a callable for token in anything long-running. Delegation tokens are TTL-bounded, so a token captured once at construction stops working the moment it outlives its TTL — and every call after that is refused. A callable is re-read per call.

Why this ships here rather than being yours to write. The deadline, the capability_version check and the error mapping all live below d.generate. A hand-rolled client would re-implement all three — and the version check is the one that tells you, in words, that your SDK is too old for the deployment you are pointed at. If you do supply your own client=, set timeout=None on it: the deadline belongs to d.generate(deadline_ms=...), and a client timeout silently pre-empts it.

✅ Core sends you the token, on every turn. It arrives in the /v1/execute body as delegation_token. Lift it before to_input (which drops unknown keys by design), bind it for the turn, and pass current_delegation_token as the token= argument:

from swisper_agent_sdk import delegation_token_from_body, use_delegation_token

with use_delegation_token(delegation_token_from_body(body)):
    result, _ = await agent.run_turn(to_input(body))

⚠ This paragraph used to say the opposite — that nothing handed a deployed agent a token. It was wrong: the SDK was discarding a credential core had always sent (D-27 Am.4).

The same token reads your prompts. http_prompts_transport(base_url, token=...) is the sibling of http_generate_transport; pass it as prompts_transport= and declare prompts in capabilities: — see "Reading the operator's edits" above. One base URL, one credential, two verbs.

What this package deliberately does not give you

Said up front, because discovering it by hitting a wall is worse:

• No swisper.* imports. The SDK never reaches into the Swisper backend, and a build check enforces that on both src/ and examples/. If something you need is missing here, that is our gap to close — tell us rather than working around it.
• No database access. Your agent never opens a database connection. Data you are entitled to arrives through a capability call that core authorises for that turn.
• A deliberately small dependency list. A test asserts that every import is declared and every declaration is imported — a property, not a count — so a new one cannot arrive as a side effect of a convenient import, undeclared. Read the list from the source rather than from this sentence, which has been wrong before: python3 -c "import importlib.metadata as m;print(*[r for r in m.requires('swisper-agent-sdk') if 'extra ==' not in r],sep=chr(10))". It was twelve on 2026-09-30: pydantic, langgraph, jsonpath-ng, pyyaml, langchain-core, httpx, jsonschema, swisper-studio-sdk, redis, jinja2, typing-extensions and cryptography (Ed25519 publisher keys; there is none in the standard library). (It has said seven, then nine, then ten, then eleven, each time correct when written. The last time it was wrong it omitted jinja2 — the dependency the whole prompts-as-files chapter rests on. Run the command; do not trust this sentence.) (See the install-time note above: langgraph alone brings a much larger dependency tree with it — the count above is what we declare, not what actually installs.) httpx backs http_generate_transport, and it is imported inside that transport's constructor rather than at module scope — so import swisper_agent_sdk stays free of anything network-capable for an agent that only runs the offline harness or verifies a delegation grant offline.
• The SDK does not configure logging. Importing it leaves your root logger untouched. Call swisper_agent_sdk.correlation.configure_logging() yourself if you want ours.