beam-directory 0.2.0

Python SDK for the Beam Protocol — SMTP for AI Agents

beam-directory · Python SDK

SMTP for AI Agents — Python SDK for agent identity, registration, discovery and intent routing via the Beam Protocol.

---

Installation

pip install beam-directory

With WebSocket support:

pip install "beam-directory[websocket]"

Quick Start

import asyncio
from beam_directory import BeamIdentity, BeamDirectory, BeamClient
from beam_directory.types import DirectoryConfig, AgentSearchQuery

async def main():
    # 1. Generate a Beam identity
    identity = BeamIdentity.generate(agent_name="myagent", org_name="myorg")
    print(f"Beam ID: {identity.beam_id}")
    # → [e-mail]

    # 2. Register with a directory
    client = BeamClient(
        identity=identity,
        directory_url="http://localhost:3100"
    )
    record = await client.register("My Agent", capabilities=["query", "answer"])
    print(f"Registered! Trust score: {record.trust_score}")

    # 3. Look up another agent
    directory = BeamDirectory(DirectoryConfig(base_url="http://localhost:3100"))
    agent = await directory.lookup("[e-mail]")
    if agent:
        print(f"Found: {agent.display_name}")

    # 4. Search agents
    agents = await directory.search(AgentSearchQuery(org="myorg", limit=10))
    for a in agents:
        print(f"  {a.beam_id} — {a.display_name}")

    # 5. Send an intent
    result = await client.send(
        to="[e-mail]",
        intent="query",
        params={"q": "What is the weather?"}
    )
    if result.success:
        print(f"Result: {result.payload}")
    else:
        print(f"Error: {result.error}")

asyncio.run(main())

Concepts

Beam ID

Every agent has a globally unique Beam ID in the format:

[e-mail]

Like an e-mail address, but for AI agents.

Intent Frames

Agents communicate via Intent Frames — small JSON objects (<1 KB) signed with Ed25519:

{
  "v": "1",
  "intent": "query",
  "from": "[e-mail]",
  "to":   "[e-mail]",
  "params": { "q": "Current pipeline status?" },
  "nonce": "550e8400-e29b-41d4-a716-446655440000",
  "timestamp": "2026-03-04T00:00:00Z",
  "signature": "<Ed25519 base64>"
}

Trust Scores

The directory assigns trust scores (0.0–1.0) based on:

• Domain ownership verification (DNS TXT record)
• Agent uptime and heartbeat frequency
• Signature verification success rate

API Reference

BeamIdentity

# Generate a new identity
identity = BeamIdentity.generate(agent_name="agent", org_name="org")

# Export / import
data = identity.export()         # BeamIdentityData
identity = BeamIdentity.from_data(data)

# Sign and verify
sig = identity.sign("payload")
ok  = BeamIdentity.verify("payload", sig, identity.public_key_base64)

# Parse a Beam ID
parts = BeamIdentity.parse_beam_id("[e-mail]")
# → {"agent": "agent", "org": "org"}

BeamDirectory

from beam_directory import BeamDirectory
from beam_directory.types import DirectoryConfig

dir = BeamDirectory(DirectoryConfig(base_url="http://localhost:3100"))

# Register
record = await dir.register(identity.to_registration("My Agent", ["query"]))

# Lookup
agent = await dir.lookup("[e-mail]")

# Search
agents = await dir.search(AgentSearchQuery(org="myorg", capabilities=["query"]))

# Heartbeat
await dir.heartbeat("[e-mail]")

BeamClient

client = BeamClient(identity=identity, directory_url="http://localhost:3100")

# Register shortcut
record = await client.register("My Agent", ["query", "answer"])

# Send intent
result = await client.send(to="[e-mail]", intent="query", params={})

# Handle incoming intents
@client.on_intent("query")
async def handle_query(frame):
    return create_result_frame(
        success=True,
        nonce=frame.nonce,
        payload={"answer": "42"}
    )

Development

git clone https://github.com/Beam-directory/beam-protocol
cd beam-protocol/packages/sdk-python

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

pytest

License

Apache 2.0 — see LICENSE.