Set up People Memory

Configure the whole path from conversation to private database. Keep credentials out of chat output, Git, browser code, and shell history where possible.

1. Inspect before asking

1. Check whether the people-memory MCP tools are already present.
2. If present, call graph_status. Offer repair or reconfiguration instead of reinstalling blindly.
3. Detect the current agent, uv, Supabase CLI, Docker-compatible runtime, and browser tools.
4. Locate a cloned people-memory-mcp repo if one exists.

Do not claim a component works because its config file exists. Verify the MCP with graph_status.

2. Ask the deployment choice

Ask one compact question with these options:

• Hosted Supabase, recommended for a free cloud backend, Auth, REST, and browser UI.
• Local Supabase, recommended when all contact data must remain on the machine.
• Existing PostgreSQL, for users who already operate a database. Explain that the UI then needs the optional API or a compatible auth/REST layer.

Read references/deployment.md for the chosen path. Do not load the other paths.

3. Set the owner and secrets

1. Store the PostgreSQL URL in ~/.config/people-memory/.env with mode 0600.
2. Generate an API token if the optional REST API will run.
3. Set app_settings.owner_email to the account that may open the hosted UI.
4. Put only the Supabase URL and publishable key in web/config.js.
5. Never put a service-role key or PostgreSQL URL in the browser.

4. Connect the current agent

Prefer the user's existing agent. Add the stdio MCP through that agent's supported command or config. For Codex and Claude Code, use the exact current commands in the repository README. Start a fresh session after changing MCP config.

Offer to install the durable memory rule from templates/AGENTS.md or templates/CLAUDE.md. Ask before changing a global user instruction file. Preserve existing content and add one clearly named section.

If the user has no agent, offer a minimal workspace containing the matching template and connect it to Codex or Claude Code. Do not build a custom agent runtime unless they ask for one.

5. Ask what to import

Ask which sources the user wants now:

• LinkedIn Connections export
• Google Contacts CSV
• WhatsApp chat export
• Another CSV for agent-guided column mapping
• Nothing yet

Invoke import-contacts for selected sources. Prefer official exports before live account access.

6. Offer optional live connectors

Ask whether the user wants to connect Gmail, Calendar, Drive, or a live WhatsApp bridge. State what each source would reveal. Use an already-authorized connector when available. Otherwise help the user connect one, but never request passwords in chat.

Offer Codex in Chrome or Claude in Chrome when available for visible Supabase setup and graph UI verification. Let the user watch and take over for login or credential entry.

Invoke enrich-people only for sources the user approves.

7. Start and verify

1. Apply migrations.
2. Start the UI locally or deploy web/ to the selected static host.
3. Call graph_status in a fresh agent session.
4. Create a synthetic test person, read it back, and delete it through delete_person or guarded SQL.
5. Open the UI and verify login, search, create, edit, and archived deletion.
6. Remove the synthetic record and its archive.

Finish with the chosen deployment, connected agents, enabled imports/connectors, UI address, and any step that still needs the user's login. Do not print secrets.