Configuration

Origin is configured through the plugin, CLI, and MCP connector. Start with /init in Claude Code or npx -y @7xuanlu/origin setup for other MCP clients.

Use origin doctor after changing setup. It checks the daemon, local runtime, MCP connector, model state, key state, and common path issues.

Spaces separate work, personal, client, and project memory without requiring multiple Origin installs. Set a space for a shell session with ORIGIN_SPACE, or define stable spaces in ~/.origin/spaces.toml.

Use origin space commands when you want the CLI to list, create, inspect, or move spaces instead of editing the file directly.

MCP clients need to launch the same local origin-mcp connector. The normal path is origin mcp add because each client stores MCP config differently.

If a client requires manual JSON, use origin mcp add --dry-run to inspect the command path for your machine, then paste the generated shape into that client's settings.

The daemon binds to 127.0.0.1:7878 by default. That is the right setting for normal laptop use because the memory API stays local to the machine.

Only change ORIGIN_BIND_ADDR when you deliberately need a non-loopback bind, such as a Docker or VM environment. Exposing the daemon is a privacy decision, not a normal setup step.

Human-facing Origin artifacts live under ~/.origin. The daemon-owned database and platform service files live under the OS application data location.

Read Markdown pages and session logs freely. Avoid hand-editing the database; use capture, review, distill, forget, and the CLI so the index, pages, and metadata stay consistent.

Origin works in local memory mode without a model download or API key. Configure a model or Anthropic key only when you want daemon-side language work such as richer extraction, recaps, or page synthesis.

Treat keys as an explicit opt-in. The connected AI client may already call its own provider during chat; Origin's model and key settings only cover daemon-side work.