apx-agent 0.3.0

Standalone agent runtime for Databricks Apps — MCP, A2A, /invocations, tool routing

apx-agent

Build governed Databricks agents. Three levels — use whichever fits:

LlmAgent	The base. You own the loop: tools, hooks, guardrails, iteration cap.
DataAgent	One line over a Unity Catalog schema. Grounded in real columns, runs as the calling user.
CoworkerAgent	Joins two source systems. Persona, join key, objective.

Deploy to Databricks Apps or Mosaic AI Model Serving — same agent, one flag.

---

LlmAgent — you control the loop

LlmAgent is an LLM + tools + a loop. You decide what it can call, when it stops, and what happens before and after each tool call. Nothing is hidden.

from apx_agent import LlmAgent, uc_function_tool, genie_tool

agent = LlmAgent(
    instructions="Investigate customer accounts.",
    tools=[
        uc_function_tool("main.tools.lookup_account"),
        genie_tool("abc123", description="Answer billing questions"),
    ],
    max_iterations=10,
    before_tool=lambda name, args: print(f"calling {name}"),
    after_tool=lambda name, result: audit_log(name, result),
)

The loop knobs: max_iterations caps runaway chains. before_tool and after_tool run synchronously around every tool call — use them for logging, redaction, rate limiting, or hard stops. input_guardrails and output_guardrails intercept at the message boundary. None of this requires subclassing.

Compose loops explicitly. LoopAgent iterates until a condition is met; SequentialAgent pipelines agents in order; ParallelAgent fans out; HandoffAgent routes conversationally. These are first-class, not hacks around a black-box framework.

from apx_agent import SequentialAgent

investigation = SequentialAgent(
    agents=[presence_check, lineage_trace, code_analysis, synthesis],
    instructions="Investigate why data is missing.",
)

See docs/workflow-patterns.md for the full composition reference.

---

DataAgent — one line over a UC schema

from apx_agent import DataAgent

agent = DataAgent("main", "sales")

That's a working agent. It knows the tables and columns in main.sales before the first question — no SHOW TABLES at runtime, no discovery prompt, no hallucinated schema.

How schema discovery works (first match wins):

1. apx scaffold bakes .apx/schema.json at project creation time by querying the UC Tables API. The manifest ships with your code, so a deployed app is grounded in the real columns with no ws= arg and no warehouse needed at startup.
2. ws=WorkspaceClient() triggers live introspection at construction time — useful when you want the freshest schema or are working outside a scaffolded project.
3. tables= explicit override — pass {"orders": ["id(bigint)", ...]} for tests or when you already have the schema from another source.
4. Ungrounded fallback — if none of the above applies, the agent falls back to generic data-assistant instructions and discovers the schema with its SQL tool on the first turn.

# Live introspection — agent knows your columns at construction time:
from databricks.sdk import WorkspaceClient
agent = DataAgent("main", "sales", ws=WorkspaceClient())

# Add a persona, Genie, vector search, or UC functions on top:
agent = DataAgent(
    "main", "sales",
    persona="a revenue analyst",
    genie_space="abc123",
    vector_index="main.sales.product_docs",
    extra_tools=[uc_function_tool("main.tools.send_alert")],
)

How governance works: deploy once, everyone runs as themselves. The app forwards each caller's OAuth token per request, and Unity Catalog enforces their grants on their data. Share the app with your team — each person queries their own slice. See docs/identity-passthrough.md.

See docs/data-agent.md for the full reference.

---

CoworkerAgent — join two source systems

Two source systems landed in a UC schema. One business entity links them. One question neither system can answer alone.

from apx_agent import CoworkerAgent

agent = CoworkerAgent(
    "main", "payroll",
    persona="a payroll operations analyst",
    join_key="employee ID",
    objective="surface mismatches between hours worked and paychecks issued",
    # memory="persistent",  # uncomment to remember facts across sessions
)

The join_key and objective are woven into the agent's grounded instructions so it always knows which field links the tables and what question it exists to answer. Common patterns:

Use case	System A	System B	Join key
Payroll reconciliation	Kronos (hours worked)	Workday (paychecks)	employee ID
Quote-to-cash	Salesforce (deals)	NetSuite (invoices)	opportunity ID
Onboarding / offboarding	Workday (employment status)	Okta (access)	employee ID
Warranty & entitlement	ServiceNow (cases)	SAP (contracts, parts)	asset serial number
Order status	Oracle ERP (orders)	TMS (freight)	PO / shipment number
Claims integrity	Epic (chart)	Claims system (coding)	patient encounter

Scaffold one:

apx scaffold my-coworker --template coworker

See docs/coworker.md for the full reference.

---

Deploy anywhere

apx deploy                                          # Databricks Apps (default)
apx deploy --target model-serving --name main.agents.my_agent

Same agent code. Same tools. Same apx.* MLflow trace schema either way. No changes to agent.py to switch targets.

---

See what you built

Every deployed agent ships with /_apx/topology — an interactive graph of agents, tools, sub-agents, and the UC / Genie / Vector Search / serving resources they reach. Click any node for its details.

See docs/dev-ui.md for the full /_apx/* surface (chat, traces, eval, tool inspector, probe).

---

📚 Docs · 🚀 Quick start · 🧪 Examples · ⚙️ CLI

---

Quick start

Python 3.11+ required.

uv add "apx-agent[langgraph]"
uv run apx scaffold my-agent
cd my-agent && uv run apx run --reload   # FastAPI on :8000, /_apx/* dev UI

When it looks right: uv run apx deploy.

Something not working? Run apx doctor — checks Python, uv, Databricks CLI, auth, and project layout. Prints a Fix: line for anything wrong.

See docs/getting-started.md for the full walkthrough including prerequisites, workspace requirements, and the local dev walk.

---

Examples

13 worked examples in python/examples/:

Example	What it shows
customer_triage	HandoffAgent + memory + UC tools
data-triage-agent	6-step SequentialAgent (presence → lineage → pipeline → genie → code → synthesis)
entity-resolution-agent	Fuzzy account match via Vector Search + HandoffAgent
memory_demo	MemoryBank + ExampleStore — recall across handoffs
voynich	LoopAgent + 5-agent evolutionary population
slack-agent	Slack-initiated runs as the Slack user's Databricks identity
+ 7 more	data-inspector, eligibility-agent, contract-parsing, shortage-intelligence, explain-my-bill, apx-builder, agent-hub

---

CLI

apx scaffold <name>                # scaffold a new agent project
apx run                            # local FastAPI dev server (/_apx/agent)
apx deploy                         # deploy to Databricks Apps (default)
apx deploy --target model-serving --name catalog.schema.agent
apx eval evalset.jsonl             # run against deployed endpoint with LLM judge
apx doctor                         # diagnose auth, deps, project layout

See docs/cli.md for the full surface.

---

Docs

Topic	Doc
DataAgent reference	docs/data-agent.md
CoworkerAgent reference	docs/coworker.md
CoworkerAgent use cases	docs/coworker-use-cases.md
Workflow patterns + composition	docs/workflow-patterns.md
Identity passthrough + OBO	docs/identity-passthrough.md
Configuration — pyproject.toml, declarative tools	docs/configuration.md
Sessions + memory + examples	docs/sessions-and-memory.md
Apps vs Model Serving	docs/apps-vs-model-serving.md
Governed primitives + UC functions	docs/governed-primitives.md
Evaluation + MLflow	docs/evaluation.md
Compliance — Watchdog, audit log, guards	docs/compliance.md
Dev UI	docs/dev-ui.md

---

For AI coding assistants

The repo ships an llms.txt index of all documentation URLs. Add the docs as a local MCP server in Claude Code:

claude mcp add apx-agent-docs --transport stdio -- \
  uvx --from mcpdoc mcpdoc \
  --urls "apxAgent:https://raw.githubusercontent.com/stuagano/apx-agent/main/llms.txt" \
  --transport stdio

---

License

Apache 2.0 — see LICENSE.