Beam ProtocolThe Trust Layer for AI Agents

Start with the question you need answered ​

Beam has one public evaluation journey. Choose the first step that matches your current question:

1. Does the trust model make sense? Open the Guided Evaluation. It explains identity, policy, trace, and operator evidence around one controlled handoff.
2. Can I see a message move? Use the Browser Playground. It creates a temporary Beam ID and shows the protocol events for one test message.
3. Can this operate one real workflow? Request a Hosted Pilot with one sender, one recipient, one owner, and one measurable outcome.
4. Am I ready to integrate? Register a Real Agent, then continue with the SDK or CLI.

Technical paths ​

• Getting Started for local setup and the first SDK flow.
• Hosted Quickstart for a self-run demonstration environment.
• Grok MCP Connector for connecting a Grok-compatible MCP client through Beam.
• Codex MCP Connector for a persistent Beam identity, contact list, and inbox in Codex.
• Identity Onboarding for Beam IDs, keys, and organization namespaces.
• Intent Lifecycle for request states, retries, and delivery semantics.
• Operator Observability for traces, alerts, audit history, and support evidence.
• Trust and Assurance for the boundary between a person, organization, agent, runtime, delegation, and payment.

Demo context ​

Some technical guides use Acme and Northwind as clearly labeled sample organizations. They are demo fixtures for reproducible examples—not customer claims and not the public product narrative.

What Beam is for ​

Beam complements agent runtimes and connectivity protocols. MCP connects an AI system to tools. A2A-style protocols move messages between agents. Beam focuses on the trust boundary: who initiated the work, what was allowed, whether human approval was required, and which evidence remains after the handoff.

Start narrow: one sender, one recipient, one intent, and one operator who needs the truth.