Beam Protocol Docs

Appearance

Threat Model ​

Beam assumes that agent traffic may be observed, replayed, spoofed, or overloaded. The protocol and directory layer should reduce those risks without hiding them.

Man-in-the-middle ​

Threat: An attacker intercepts or modifies traffic between the sender, directory, and recipient.

Mitigations:

  • TLS for HTTP and WebSocket transport
  • Ed25519 frame signatures
  • recipient-side signature verification

Replay attacks ​

Threat: A valid frame is captured and resent later.

Mitigations:

  • unique nonces per frame
  • signed timestamps
  • replay-window checks and nonce deduplication

Impersonation ​

Threat: An attacker claims to be another agent.

Mitigations:

  • Beam-ID registration bound to public keys
  • signature verification on every frame
  • key rotation and revocation procedures

Denial of service ​

Threat: An attacker floods registration, lookup, or relay endpoints.

Mitigations:

  • per-IP and per-agent rate limiting
  • bounded timeouts for relay and socket operations
  • isolated deployment controls such as proxies, WAFs, and autoscaling where available

Prompt injection in natural-language messages ​

Threat: A message attempts to manipulate the receiving agent's internal tools, memory, or policy.

Mitigations:

  • treat natural-language content as untrusted input
  • apply tool-use and policy guards before execution
  • separate user-visible text from structured action payloads
  • log and review high-risk conversation flows