Beam Protocol Docs

Appearance

Core Concepts ​

Beam Protocol is a small set of building blocks for secure agent-to-agent communication.

Beam-ID ​

A Beam-ID is the global address for an agent, formatted like an email address:

text
agent@org.beam.directory

It identifies both the agent name and the organization namespace.

Intent Frames ​

Intent frames are request messages. They usually include:

  • protocol version
  • sender Beam-ID
  • recipient Beam-ID
  • intent name
  • payload or params
  • nonce
  • timestamp
  • Ed25519 signature

Use intent frames for structured operations such as search.query, workflow.start, or conversation.message.

Result Frames ​

Result frames are replies to intents. They normally include:

  • the original nonce for correlation
  • success status
  • a response payload
  • optional error and error code
  • latency metadata
  • a signature from the responding agent

Together, intent and result frames create a verifiable request-response protocol for agents.

Directory ​

The directory is the shared coordination layer for Beam. It handles:

  • agent registration
  • discovery and lookup
  • intent relay
  • WebSocket fan-in and fan-out
  • operational health signals

You can run one public directory, a private team directory, or multiple federated directories.

Trust Scores ​

Trust scores help agents rank or filter potential peers. A directory may calculate trust from signals like:

  • successful registrations
  • uptime and heartbeat freshness
  • delivery success rate
  • verification or policy status

Trust scores are advisory, not absolute. Agents should still verify signatures and enforce local policy.

ACL ​

Access control lists define who may send which intents to which targets.

Typical ACL rules answer questions like:

  • which Beam-IDs may call a sensitive intent
  • whether a wildcard sender is allowed
  • which workflows are internal-only

ACLs are a policy layer on top of cryptographic identity.