Skip to main content

Open Agent Protocol — a routing layer for inter-agent task handoff

Project description

OAP — Open Agent Protocol

A lightweight routing layer for passing tasks between AI agents.

OAP defines a standard envelope format (TaskEnvelope) and a router that dispatches tasks to the right agent based on capabilities or explicit handoff instructions.

Installation

pip install open-agent-protocol

Quick start

from oap import TaskEnvelope, OAPRouter
from oap.adapters.http import HTTPAdapter

router = OAPRouter()
router.register(
    "research-agent",
    HTTPAdapter(agent_id="research-agent", base_url="http://localhost:9000"),
    capabilities=["research", "search", "find"],
)

envelope = TaskEnvelope(goal="research the best vector databases")
result = await router.route(envelope)
print(result.memory["last_result"])

CLI

# Create a new task envelope
oap init "research the best vector databases" --output task.json

# Register an HTTP agent — OAP discovers capabilities automatically from GET /
oap register research-agent http://localhost:9000
# Falls back to manual capabilities if the agent has no GET / endpoint
oap register research-agent http://localhost:9000 --capabilities "research,search,find"

# Route the envelope to the best matching agent
oap route task.json --output result.json

# Automatically follow handoffs until the task is complete
oap chain task.json --output final.json

# List all interrupted chains (saved mid-flight)
oap runs

# Resume a chain that failed or was interrupted
oap resume <envelope-id>
oap resume <envelope-id> --pipeline "agent-a,agent-b,agent-c"

# Inspect an envelope
oap inspect result.json

# Validate envelope structure
oap validate result.json

# List all registered agents
oap agents

# Check reachability of all registered agents
oap ping

# Remove an agent from the registry
oap unregister research-agent

Chaining agents

When an agent sets a handoff.next_agent on its response, oap chain automatically routes to the next agent and keeps going until the task is complete or a hop limit is reached.

# Python API
result, visited = await router.chain(envelope, max_hops=10)
print(" → ".join(visited))  # e.g. research-agent → summarise-agent
# CLI — follows handoffs automatically, prints each hop
oap chain task.json --output final.json --max-hops 5

The chain stops when:

  • An agent returns a response with no handoff set, or
  • max_hops is reached (default: 10)

Resuming failed chains

If a chain is interrupted (network error, agent down, hop limit hit), OAP saves progress after each hop. Resume from the last successful hop with:

oap runs                        # list all interrupted chains
oap resume <envelope-id>        # resume handoff-driven chain
oap resume <envelope-id> --pipeline "agent-a,agent-b,agent-c"  # resume a fixed pipeline
oap resume <envelope-id> --max-hops 5  # cap additional hops

Progress is stored in ~/.oap/runs/ and deleted automatically when the chain completes.

Registry

Agents are stored in ~/.oap/agents.json and persist across commands.

When an agent implements GET / returning {agent_id, capabilities, description}, registration is automatic:

oap register my-agent http://localhost:9000
# → OAP hits GET /, reads capabilities and description, saves everything

oap register my-agent http://localhost:9000 --capabilities "research,find"
# → fallback: use provided capabilities if GET / is unavailable

oap agents          # list all registered agents with description column
oap ping            # health-check all agents, auto-updates capabilities if changed
oap unregister my-agent

Agent health endpoint

Add GET / to your agent to enable self-registration and health checks:

@app.get("/")
async def info():
    return {
        "agent_id": "my-agent",
        "capabilities": ["research", "find", "search"],
        "description": "Researches topics and returns structured findings.",
        "status": "ok",
    }

Concepts

  • TaskEnvelope — the standard task object passed between agents. Contains the goal, memory, steps taken, and optional constraints.
  • OAPRouter — selects the best registered agent for a given envelope and invokes it.
  • AgentAdapter — translates between the envelope format and an agent's native interface.
  • HTTPAdapter — built-in adapter for agents that expose a POST /invoke endpoint.

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

open_agent_protocol-0.2.5.tar.gz (275.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

open_agent_protocol-0.2.5-py3-none-any.whl (17.0 kB view details)

Uploaded Python 3

File details

Details for the file open_agent_protocol-0.2.5.tar.gz.

File metadata

  • Download URL: open_agent_protocol-0.2.5.tar.gz
  • Upload date:
  • Size: 275.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for open_agent_protocol-0.2.5.tar.gz
Algorithm Hash digest
SHA256 2a70f7d7e828cbe8949130eeb177f20844f5068b5bc78978bfaf118079ce1738
MD5 6c2007744a1c25d3972b0467a623d5de
BLAKE2b-256 9b9e5c574ef55ae18692b91b93b01777c18de87bb53644b6a2b15f5d7ca0ca51

See more details on using hashes here.

File details

Details for the file open_agent_protocol-0.2.5-py3-none-any.whl.

File metadata

File hashes

Hashes for open_agent_protocol-0.2.5-py3-none-any.whl
Algorithm Hash digest
SHA256 4189be47b4b587e1a38b6d5c79a4596c56b483acc716c501da721adcda525fe1
MD5 2f58ad502dcd523726c4fd7a1fbad12e
BLAKE2b-256 515e08b238e3a4ae6e769f887a2470c8e50275dc5508f89fcfdb0ef5d68d1431

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page