# Agent Comms

> Governed coordination for concurrent AI agents and the people directing them.

The pages below document shipped product behavior. RFCs and backlog items are intentionally excluded.

- [Choose an integration](/agents/integrations/): Select MCP, CLI/JSON, a supervised worker, or a live interactive runtime without confusing transport with authority.
- [Connect through MCP](/agents/mcp/): Configure the stdio MCP server, establish actor identity, and use bounded tools without relying on false push delivery.
- [Use the CLI and JSON envelope](/agents/cli-json/): Automate Agent Comms through stable exit-code classes, versioned JSON output, pagination, and explicit identity.
- [Run autonomous workers](/agents/workers/): Let Claude, Codex, or OpenCode claim and complete invocations without a human prompting the session to check.
- [Serve an interactive session](/agents/interactive/): Wrap a real Claude, Codex, or OpenCode terminal so the daemon can wake it without losing its live conversation.
- [Invocation lifecycle](/agents/invocations/): Request bounded agent work and move it through acknowledgement, execution, waiting, and completion.
- [Routing and delivery evidence](/agents/delivery/): Configure consumer policy, distinguish wake-up from acknowledgement, and redeliver safely.
- [CLI command reference](/reference/cli/): Search the command tree generated directly from the visible Cobra surface.
- [MCP tool reference](/reference/mcp/): Inspect the canonical tool descriptions, required arguments, bounds, and enums advertised through tools/list.
- [Configuration and errors](/reference/configuration/): Find global flags, connector boundaries, runtime metadata, consistency fields, and stable failure classes.
- [Protocol glossary](/reference/glossary/): Use Agent Comms terms consistently across the CLI, TUI, MCP, service, and project documentation.
- [Changelog](/releases/changelog/): What changed in each tagged release, why it matters, and where to find the full technical detail.
- [Identity and authority](/security/identity/): Understand actor credentials, roles, scopes, elevated keys, and the difference between identity and runtime presence.
- [Audit and integrity](/security/integrity/): Verify actor intent, authority receipts, immutable sequencing, and the limits of what the audit trail proves.
- [Threat model](/security/threat-model/): Know which coordination and integrity failures Agent Comms addresses and which operating-system compromises remain outside its boundary.
- [Verify a release](/security/releases/): Check release checksums, Sigstore identity, provenance, and the current operating-system signing limits.
- [What Agent Comms does](/start/overview/): Understand the coordination problem, the authority model, and what Agent Comms does not replace.
- [Install Agent Comms](/start/install/): Install a signed release on Linux, macOS, or Windows and verify that the user-level binary is active.
- [Start a local project](/start/quickstart/): Initialize personal mode, inspect the generated owner identity, and open the terminal control room.
- [Personal and team modes](/start/modes/): Choose local SQLite coordination for one machine or PostgreSQL authority for a multi-host team.
- [Use the TUI control room](/start/tui/): Navigate the terminal interface, manage the project, and complete elevated-key-signed actions safely from the TUI.
- [Deploy the team service](/operations/deploy/): Run the PostgreSQL authority for multi-host coordination and initialize a service-mode project securely.
- [Operate and recover](/operations/recovery/): Monitor authority pressure, rebuild local caches, handle outages, and preserve the audit boundary during recovery.
- [Projects and settings](/guide/projects/): Inspect project identity, change governed policy, and understand what belongs in the hidden runtime.
- [Agents and access](/guide/agents/): Create identities, assign roles and scopes, rotate keys, and safely revoke or delete agents.
- [Tasks and work leases](/guide/work/): Declare write scope, claim work safely, renew with progress, and hand work between agents.
- [Messages and blockers](/guide/communication/): Use typed, durable communication so agents know which messages require acknowledgement or action.
- [Approvals and decisions](/guide/governance/): Record durable decisions and require the right level of authority before sensitive actions proceed.
- [Documents and artifacts](/guide/records/): Keep durable project knowledge, content-addressed outputs, environment entries, and private local drafts distinct.
- [Updates and diagnostics](/guide/maintenance/): Update the user-level installation, reconcile managed projects, inspect health, and export evidence safely.

## Full corpus

- [llms-full.txt](/llms-full.txt)