Bring context into the conversation
Ask what you decided, what happened recently, or what you know about a person or project.
Geppetto works best when your AI recalls context before answering, captures useful facts as you work, and keeps open work in one shared task list. You can use the same memory from every connected client.
After signing in, follow the installation steps for ChatGPT, Claude, Cursor, or your preferred MCP client. Then start a new conversation and ask:
Your new vault already contains onboarding tasks. The AI can walk through them with you, so you do not need to configure everything before your first conversation.
Ask what you decided, what happened recently, or what you know about a person or project.
Say “remember this” for an important fact, preference, decision, or relationship you want available later.
Ask Geppetto to create reminders and follow-ups instead of leaving them in one client’s private task system.
At the start or end of a session, ask for open tasks and update what has changed.
Geppetto already sends its rules through MCP. This short instruction is optional, but useful in clients that support persistent personal instructions.
Optional instruction
Geppetto Memory is connected. Search it before asking me about my life, work, people, or preferences. After every message I send, identify and save new durable facts about my life, work, people, and projects, plus decisions, commitments, and follow-ups, without waiting to be asked. Resolve and update an existing canonical note before creating a new one. Keep reminders and tasks there so every AI I use sees them. For how to work in it, memory_read 50 Knowledge/Geppetto Memory.md.App description, when a client asks for one
My long-term memory. Search it before asking me about people, organizations, projects, decisions, preferences, or commitments. After every user message, identify and save new durable facts about my life, work, people, and projects, plus decisions, commitments, and follow-ups, unasked. Resolve and update an existing canonical note before creating a new one. Keep reminders, follow-ups, and open work as tasks here rather than in this app alone, so every AI I use sees the same list. Run the daily check-in when it is due. Nothing is ever deleted.Add the instruction under Settings → Personalization → Custom Instructions. Use the app description in the connector’s Description field.
Add the instruction under Preferences or to a project instruction.
Use it only when the client has a persistent instructions field. Otherwise no extra setup is needed.
A check-in can review the trusted sources you choose, reconcile new requests with your tasks, and prepare a short daily overview. Run the first check-in in an open conversation so you can verify that the client can reach every source.
Decide which mailboxes, calendars, issue tools, and chat channels the AI may review.
Keep the conversation open while the AI reviews the sources and confirms what it saved. Source setup alone does not perform a check-in.
A capable AI client may create one recurring task. Prefer a cloud schedule for early morning (for example 06:00). If the schedule runs on your computer, pick a time when it is usually awake — around 09:00 works better than 06:00. Geppetto cannot wake the client or access Gmail, Slack, calendars, or other sources by itself. Every client shares a completed result, so only one schedule is needed.
Journal what happened, update existing notes when durable state changes, and create tasks for follow-ups. Compare yesterday’s calendar with the vault. Do not change mail, issues, or chat without showing the exact proposal first.
A search hit is a clue, not a status. The AI should open the thread or calendar event, look past the checkpoint for the reply that answers it, and say what proves each status before telling you something is unanswered.
Friday's launch date was confirmed. The launch project was updated. New person: [[Jacob]], product lead at Northwind.
Tasks created or closedCreated “Send [[Jacob]] the final launch plan”, due Thursday.
Needs your inputTwo possible owners were mentioned for the release checklist. Who owns it? Slack was unreachable, so that source was skipped.
Tell the AI what is wrong and ask it to update the memory. Geppetto keeps history rather than silently deleting information.
Do not save passwords, API keys, access tokens, or information whose disclosure could cause serious harm.
Connect each AI to the same account. Facts and open tasks then follow you between conversations and clients.
Markdown content remains readable while SQLite provides transactional storage and search.