Operate

Running locally

The daemon, corpus and MCP server run entirely on your machine with no relay. What that covers today, and where self-hosting the relay stands.

Where self-hosting stands today

Being straight with you: you cannot stand up your own relay yet. The relay’s source is not public, so there is nothing for you to build or deploy. A self-hosting guide is planned, and this page will carry it when it lands.

What is available: the CLI, the corpus and the MCP server are source-available under the Business Source Licence 1.1 (Change Date 2030-05-19, then Apache 2.0). That covers everything that runs on your own machine.

What you can run locally, right now

This is not a consolation prize — it is a real mode, and it is the one most people want. Without any relay at all:

• Harvesters keep running and writing to ~/.dotperson/corpus.db.
• The embedder keeps indexing, so hybrid search works.
• The synthesis pass runs.
• The MCP server answers read_persona, corpus_query, corpus_recent and friends from local state.
• personkit today, personkit corpus query, personkit corpus surfaces and personkit export are all local — no network needed.

personkit daemon --install   # local-first cycle: harvest → embed → synthesise

If personkit login has never run, the daemon’s observe and sync stages early-return — no credentials means no POST. Nothing is sent anywhere.

What you give up by staying local: cross-device sync, browser-extension capture (the extension posts to the relay, so it needs one), and cloud synthesis.

Pointing the CLI at a different relay

The plumbing for a custom relay URL already exists, so it will work the moment a relay is something you can run. Pass it to setup or login:

personkit setup --relay https://your-relay.example.com

It is persisted into ~/.dotperson/credentials.json as relay_url, and every later command reads it from there — there is no relay environment variable to export. The daemon reads the stored URL on each cycle, so its harvest → embed → observe → sync loop follows it automatically. (The launchd plist itself only exports HOME and PATH; the Linux systemd unit takes the user manager’s environment.) To switch later, re-run personkit login --relay https://new-relay.example.com; the next cycle picks it up.

What the relay does

For context on what a self-hosted relay would need to replace, its jobs are narrow:

• Firebase auth and session handling.
• GitHub App brokerage for your persona repo.
• Persona synthesis from captured observations (Google Vertex AI).
• Cross-device sync: receiving mirrorable corpus events, browser-extension captures and IDE observer sweeps.

Everything that makes aiperson useful day to day — the corpus, the persona cache, retrieval, the daily brief — is local.