Agent Profiles

An agent profile represents a writer or client that uses the local daemon: Claude Code, Codex, Cursor, Claude Desktop, Gemini CLI, or a local script.

Profiles make source attribution explicit. When memory has a source_agent value, later review and diagnostics can show where it came from.

Use the CLI when you want to see which agents Origin knows about or inspect one profile in detail.

This is useful after connecting a new MCP client, debugging an unexpected capture source, or checking whether a client is enabled.

Trust is a signal for how Origin should treat a known writer; it is not a replacement for reviewing uncertain memory.

Enabled state is an operational switch for a profile. If a client is misconfigured or writing noisy captures, inspect the profile before deleting memory.

Agent profiles answer who wrote this. Spaces answer which work context this belongs to.

Keep both dimensions. A trusted Claude Code profile can still write into the wrong space if the workspace resolver or ORIGIN_SPACE is wrong.

Most users do not need to edit profiles during normal daily work. The profile page matters when multiple MCP clients write to the same store, when a script imports memory, or when review shows an unfamiliar source.

For normal setup, connect the client first, run doctor, then capture and recall one harmless fact to confirm the route.