Nexus Mesh — Python SDK

The official Python SDK for Nexus Protocol, the universal open-source coordination protocol for AI agents.

Nexus lets agents — regardless of language, framework, or vendor — discover each other, communicate with end-to-end encryption, and collaborate on distributed tasks.

pip install nexus-mesh

The distribution is named nexus-mesh; the import module is nexus_sdk.

Quick start

Start the coordination hub:

nexus hub

Write a worker agent:

import asyncio
from nexus_sdk import NexusAgent

async def compute(input_data, task):
    return {"result": input_data["a"] + input_data["b"]}

async def main():
    agent = NexusAgent(
        name="calc_bot",
        capabilities=["calculate"],
        roles=["worker"],
    )
    agent.on_task(compute)
    await agent.connect()
    await asyncio.Future()   # stay online

asyncio.run(main())

Delegate work from an orchestrator:

from nexus_sdk import NexusAgent

orchestrator = NexusAgent(name="lead", roles=["admin"])
await orchestrator.connect()

# Find an agent by capability, then hand it a task
found = await orchestrator.discover(capabilities=["calculate"])
result = await orchestrator.submit_task(
    title="Add two numbers",
    assignee=found["agents"][0]["name"],
    input_data={"a": 20, "b": 22},
)
# {"result": 42}  — encrypted end-to-end in transit

Bridge an existing framework agent

Wrap a LangChain, CrewAI, AutoGen, or LlamaIndex agent without changing a line of it — it becomes discoverable and receives delegated tasks like a native Nexus agent:

from nexus_sdk.adapters.langchain import NexusLangChainAdapter

agent = NexusLangChainAdapter(my_agent_executor, name="analyst", capabilities=["market_analysis"])
await agent.connect()

Orchestrate multiple agents

from nexus_sdk import NexusPipeline

pipeline = (
    NexusPipeline(orchestrator)
    .step("Translate", capabilities=["translate"])
    .step("Calculate", capabilities=["calculate"],
          input_fn=lambda prev: {"expression": prev["translated_text"]})
)
result = await pipeline.run({"text": "compute forty two doubled"})

fan_out(orchestrator, branches, capabilities=...) runs independent branches in parallel instead. Full guide: docs/AGENT-INTEGRATION.md.

Features

• End-to-end encryption — RSA-2048-OAEP + AES-256-GCM. The hub routes ciphertext it cannot read.
• Verifiable identity — SHA-256 fingerprints over roles, permissions, and capabilities.
• JWT authentication — every message after registration carries a hub-signed token.
• Role-based access control — per-agent policies enforced at the hub.
• Discovery — locate agents by capability, role, metadata, or name.
• Distributed tasks — async lifecycle: pending → running → completed / failed.
• Federation — hub-to-hub peering across organizations, encryption preserved end to end.
• Immutable audit log — Merkle-chained events; retroactive edits break the chain.
• Rate limiting — token-bucket throttling per agent.
• Developer CLI — nexus hub | discover | ping | ask | task | keygen | dashboard | docs.

API summary

Full reference: docs/API-REFERENCE.md

Documentation

• Agent integration — adapters and orchestration
• RFC-001 — Core protocol specification
• Security and encryption model
• API reference

License

Apache 2.0