Docs
Personal Memory
One memory for all your AIs. Teach it once on Staraze; Claude, Cursor and ChatGPT all read from and write to the same place.
What it is
Personal Memory is a private, growing store of what matters to you: who you are, your projects, decisions you made, open todos, preferences. It lives on Staraze — not inside any one AI app — so switching from Claude to Cursor to ChatGPT doesn't mean starting over.
- Project-scoped. Each project has its own notes and a living summary, so your AI loads only what is relevant.
- Cross-tool. The same memory is reachable from every MCP-capable client with one key.
- Pay per use. Reads and saves are billed per token from your wallet, with the cost shown on every call. You pick the level; we never silently cut quality to save money.
Start at /memory: create your personal AI, train it, and create your key.
Privacy & ownership
- It's your data. Only your account can read or change it. Memory is never shown to other users and never used to train other personas.
- What we store: the items extracted from what you save (facts, decisions, todos, preferences) and a short summary per project. Not full conversation transcripts.
- Export any time. Settings has a full export of every item and summary as Markdown and JSON (once per 24 hours, free).
- Delete any time. Remove an item in the app, or ask your AI to
forgetit. - Your key is a password. It's shown once; rotate or revoke it any time.
Connect your AI
Create your pk_ key at /memory (it is shown once), then use the block for your client. Replace pk_your_key_here.
// claude_desktop_config.json (Settings → Developer → Edit Config)
{
"mcpServers": {
"staraze": {
"command": "npx",
"args": ["-y", "staraze-mcp"],
"env": {
"STARAZE_API_KEY": "pk_your_key_here"
}
}
}
}Grok and other apps without MCP: open /memory, choose a level, and use Generate & copy to copy your context brief, then paste it at the start of the chat. It is a snapshot — nothing is saved back automatically.
Tip: say “load my memory for project acme” at the start, and “remember that we chose Postgres” when something is decided.
Tool reference
Memory tools require your own personal key. A key for someone else's persona returns MEMORY_NOT_AVAILABLE. Every response ends with a usage line: tokens, dollars and your wallet balance.
| Tool | Arguments | What it does | Cost |
|---|---|---|---|
get_context | project?, task?, level?, cursor? | Loads your memory brief: profile, project summary, relevant notes and open todos. Levels: minimal | medium (default) | full (paged). | By level |
remember | text, project?, mode? | Saves to memory. Modes: quick | standard (default) | detailed. Duplicates are merged and outdated items superseded. | By mode |
search_memory | query, project?, limit? | Semantic search; returns items with ids. | Minimal read |
forget | item_id | Deletes one item by id. | Free |
set_active_project | project | Sets the project used when none is given. Creates it if new. | Free |
list_projects | — | Lists projects with counts and the active one. | Free |
get_pricing | — | Returns the live pricing table. | Free |
ask_persona | message | Chat with your personal persona. | Per token (API rates) |
Levels & costs
These are the defaults; Staraze sets them in its admin settings and get_pricing always returns the live numbers. Costs are estimates for the Staraze side only — your AI client also uses these tokens in its own context window.
| Reading (get_context) | Token budget | Est. cost |
|---|---|---|
minimal | 600 | ≈ $0.001 |
medium (default) | 2,000 | ≈ $0.004 |
full | 6,000 per page (paged, never cut) | ≈ $0.012 per page |
| Saving (remember) | What happens | Est. cost |
|---|---|---|
quick | Stored as you wrote it | ≈ $0.0005 |
standard (default) | Extract, de-duplicate, supersede | ≈ $0.002 |
detailed | Standard + immediate summary rewrite | ≈ $0.006 |
Empty wallet? Calls return WALLET_EMPTY with a link to add funds.
How it stays tidy
- Dedupe. When a new note is nearly identical to an existing one, the existing item is refreshed instead of adding a copy.
- Supersede. If you say “we switched from X to Y”, the old statement is marked superseded (kept in history, left out of briefs) and the new one becomes current.
- Projects. Notes belong to a project. Each project has a living summary that is rewritten, not appended, so it stays short. Use
set_active_projectso you don't repeat the name. - Types. profile, fact, decision, todo, preference, event, snippet. Profile items are always loaded first.
FAQ
Can I just paste it into Claude?
Yes — copy a context brief from /memory and paste it anywhere. But a pasted brief is a frozen snapshot for one chat in one app. The product is the live memory: it updates as you work, is scoped per project, de-duplicates itself, and follows you across Claude, Cursor and ChatGPT without copy-paste.
What gets saved?
Only what you tell your AI to remember (or add in the app), as extracted items plus project summaries. Not whole conversations.
Why is the cost shown on every call?
So you choose. Pick minimal for a quick nudge, full when you need everything; we never trim a level behind your back.
Can others see my memory or use my key?
No. Memory is private to your account. Keep your key secret; if it leaks, revoke it and create a new one.
How do I leave?
Export everything from Settings and delete your items. Your data is yours.
Looking for creator personas?
See the API & MCP reference.