naylence-agent-sdk 0.4.1

Naylence Agent SDK

Naylence Agent SDK (Python)

The Naylence Agent SDK is the official toolkit for building agents and clients on the Naylence Agentic Fabric. It gives you a clean, typed, async-first API for composing tasks, streaming results, and wiring agents together—locally or across a distributed fabric.

If you're new to Naylence, start here. For lower‑level transport/fabric internals, see naylence‑runtime.

---

Highlights

• Ergonomic agent model — subclass BaseAgent (one-shot) or BackgroundTaskAgent (long‑running/streaming) and focus on your logic.
• Typed messages & tasks — Pydantic models for Task, Message, Artifact, and JSON‑RPC A2A operations.
• Async all the way — non‑blocking lifecycle with easy scatter‑gather helpers (Agent.broadcast, Agent.run_many).
• Remote proxies — call agents by address or capabilities via Agent.remote_* helpers.
• Streaming & cancellation — subscribe to live status/artifacts; cancel in‑flight work.
• FastAPI integration — drop‐in JSON‑RPC router (create_agent_router) and /agent.json metadata endpoint.
• Security ready — works with runtime security profiles; strict‑overlay requires the naylence‑advanced‑security add‑on.

---

Install

pip install naylence-agent-sdk

Python 3.12+ is required.

---

Quickstart (minimal)

import asyncio
from typing import Any
from naylence.fame.core import FameFabric
from naylence.agent import Agent, BaseAgent

class EchoAgent(BaseAgent):
    async def run_task(self, payload: Any, id: Any) -> Any:
        return payload

async def main():
    async with FameFabric.create() as fabric:
        address = await fabric.serve(EchoAgent())
        echo = Agent.remote_by_address(address)
        print(await echo.run_task("Hello, world!", None))

asyncio.run(main())

For a gentle, runnable tour—from single‑process to distributed orchestration—use the Examples repo: https://github.com/naylence/naylence-examples-python.

---

Core concepts

Agents & tasks

• Implement run_task(payload, id) for simple one‑shot work, or override start_task(...)/get_task_status(...) for background jobs.
• Message.parts carries either text (TextPart) or structured data (DataPart).
• Long‑running flows stream TaskStatusUpdateEvent and TaskArtifactUpdateEvent until terminal (COMPLETED/FAILED/CANCELED).

Remote proxies

• Agent.remote_by_address("[e-mail]") to call a known address.
• Agent.remote_by_capabilities(["agent"]) to call by capability (fabric does resolution).

Streaming & cancel

• subscribe_to_task_updates(...) yields status/artifacts live.
• cancel_task(...) requests cooperative cancellation when supported by the agent.

RPC operations

• A2A JSON‑RPC methods (tasks/send, .../get, .../cancel, etc.) are provided for task lifecycle.
• Custom functions can be exposed via the RPC mixin in the underlying fabric (e.g., streaming operations).

FastAPI router

• Use create_agent_router(agent) to expose a JSON‑RPC endpoint (default: /fame/v1/jsonrpc) and GET /agent.json to return an AgentCard.

---

Choosing an agent base class

• BaseAgent — great for synchronous/short tasks; the default fallback packages your return value into a Task(COMPLETED).
• BackgroundTaskAgent — best for long‑running/streaming work. You implement run_background_task(...); the base manages queues, TTLs, and end‑of‑stream.

Both base classes include sensible defaults (poll‑based streaming, simple auth pass‑through). You can override any part of the lifecycle.

---

Development workflow

• Add your agents in a project with the SDK.
• Use FameFabric.create() in tests or local scripts to host agents in‑process.
• For distributed setups, operate a sentinel/fabric with naylence‑runtime (or your infra) and connect agents remotely.
• Use the Examples repo (https://github.com/naylence/naylence-examples-python) to learn patterns like scatter‑gather, RPC streaming, cancellation, and security tiers.

---

API Documentation

To generate API documentation from source:

# Generate Markdown docs to docs/api/
poetry run docs-gen

# Or specify a custom output directory
poetry run docs-gen --out path/to/output

The generated docs are Markdown files compatible with Nextra and other documentation systems.

---

Security notes

The SDK runs on the Naylence fabric’s security profiles:

• direct / gated / overlay modes work out‑of‑the‑box with the open‑source stack.
• strict‑overlay (sealed overlay encryption + SPIFFE/X.509 identities) is available only with the naylence‑advanced‑security package.

See repo links below for the advanced add‑on and images that bundle it.

---

Links

• Agent SDK (this repo): https://github.com/naylence/naylence-agent-sdk-python
• Examples (Python): https://github.com/naylence/naylence-examples-python
• Runtime (fabric & transports): https://github.com/naylence/naylence-runtime-python
• Advanced Security add‑on: https://github.com/naylence/naylence-advanced-security-python

Docker images:

• OSS: naylence/agent-sdk-python
• Advanced: naylence/agent-sdk-adv-python (includes naylence-advanced-security; BSL-licensed add-on)

---

License & support

• License: Apache‑2.0 (SDK).
• Issues: please use the appropriate GitHub repo (SDK, Runtime, Examples, Advanced Security).