Finally, every AI knows you.
Persnally learns who you are from your AI history — your chats, your code, your decisions — and serves it to every AI you use. On your machine. Yours.
what you're intolocal-first systems · weight 0.95how you decide“hand-roll it, not a new dep”how you writeterse · lowercase · no emojiyour conventionspnpm, never npmwhy it knowslearned from event #412 · deletableOne command connects them all · ChatGPT history imports too
You explain yourself. Again. And again.
ChatGPT doesn't know what you told Claude. Your coding agent relearns your stack every session. Each tool meets you as a stranger — and the vendors can't fix it, because their business is keeping you inside their walls.
Claude sees
ChatGPT sees
Cursor sees
“My stack: TypeScript, Next.js, Tailwind.”
Your context, in every tool you touch.
Import your history
One command finds your Claude & ChatGPT exports, your Claude Code sessions, and your git repos, and reads them.
$ persnally import
- Claude & ChatGPT exports
- Claude Code sessions
- git repos
It learns, on your machine
A local daemon turns that history into an evidence-linked model of you — never our cloud.
building your model…
- ships the smallest design
- prefers SQLite
- guards user trust
on your machine
Every AI reads it
Over MCP — the protocol your tools already speak. Claude Code pulls it in the moment a session starts; Claude, Cursor, and your agents read it whenever they need you.
Claude · session started
↳ loaded your context
persnally_context 2ms
Not a metaphor. The actual dashboard.
Every engraving on this page is standing in for something real. Below is the real thing, live — the same page served at localhost:4983 on your machine. Scroll it, click a node, judge an answer. It's running in preview mode: sample data, nothing writable, nothing sent anywhere.
The only context engine your AI can ask.
Other tools store facts. Persnally answers what you'd do — with a confidence score, deferring to you when the evidence is thin.
It answers, or it defers
Your agent asks Persnally instead of interrupting you. Below its confidence bar, it sends the agent back to you — never a made-up answer.
persnally_ask “tests before I merge?”
↳ yes — you demand proof before merge; add e2e on the changed paths.
confidence 0.92 · 3 evidence events
Correct it once, it sticks
Tell it it's wrong and the correction becomes authoritative — it outranks everything inferred, and the wrong answer never comes back.
persnally correct “I use pnpm, not npm”
↳ recorded — authoritative.
every future answer respects it
Same prompt — “write a Slack message telling my team the deploy went out” — asked of a blank AI and an AI that read your Persnally:
Generic AI
🚀 Deploy is live! Just pushed the latest changes to production. Everything looks good so far — let me know if you spot anything unexpected.
AI + Persnally
deploy is out. watching sentry/logs for the next hour, will confirm clean or flag issues here.
Terse, lowercase, no emoji, and it knew to watch Sentry — because that's how you actually work. Not configured; learned.
More than memory. An engine.
Structured events, derived views, a walkable provenance graph — decay-aware, and entirely your own.
One context, every tool reads it.
speaks MCP — adopted by every major AI
Every claim cites its evidence.
Guards user trust as non-negotiable.
↳ why does it think this? 3 events
- #128 · imported 142 Claude conversations
- #412 · vetoed telemetry without consent
- #087 · chose local-first storage
3 events · 0 guesses · each one deletable
Decide exactly what each AI can see.
Everyone remembers. Persnally answers.
No one else answers what you'd do, learns when you correct it, and lets you verify all of it on your own machine.
| Capability | Persnally | claude-mem | Mem0 · Supermemory | Unabyss · cloud vaults |
|---|---|---|---|---|
| Your AI can ask it — answer with a confidence score, defers when unsure | ||||
| Learns when you correct it — the correction outranks everything inferred | ||||
| Synthesizes a model of you — taste, voice, how you decide | ||||
| Local plaintext on your machine + source you can audit | ||||
| Reads your chat exports + git history |
has it · partial · no. Verified against each product's own docs and repos, July 2026.
A context engine you can actually trust.
Trust isn't a privacy policy here — it's the architecture. Not promises; properties you can check.
Local-first
Your context lives in ~/.persnally — not our cloud, not any vendor's silo.
~/.persnally
- persnally.db
- config.json
stays on your machine, never our cloud
Truly deletable
Forget a topic and everything derived from it is erased, then rebuilt.
$ persnally forget “rust”
- ✓ 18 events erased
- ✓ derived views rebuilt
no tombstones · no residue
Provenance-complete
“Why does it think this?” is a real lookup, never a guess.
↳ why “guards user trust”?
- #412 · vetoed telemetry
- #087 · chose local-first
- #203 · removed analytics
3 events · 0 guesses
Source-available
Read the engine, audit the claims, run it yourself. The schema and MCP interface are an open spec.
Every AI knows you. And it's yours.
Every other memory tool
“a place to store facts about you”
- A database of you — you save, you search, you retrieve
- Recalls facts; can't answer what you'd decide
- Correct it? Delete a memory, at best
- A black box — it can't tell you why
Persnally
your own context engine
- A model of you your AI can ask — with a confidence score
- Answers what you'd do; defers to you when unsure
- Correct it once and it sticks — outranks everything inferred
- Every claim cites its evidence — local, auditable, deletable
The engine is free. Forever.
Free · Pro · Teams later
Everything that touches your data runs on your machine, free. Pro adds cloud services on top — and tells you exactly which of them can see what.
Free
$0 forever
- The full engine — import, learn, synthesize your profile
- Serve your context to every AI over MCP
- The dashboard: inspect, audit provenance, delete
- Bring your own key, or run fully local with Ollama
Pro
Coming soon$6 / month · billed annually · founding price
- Encrypted backup & restore of your accumulated self — ciphertext, we hold no key
- Agent relay — your phone and cloud agents reach your context, end-to-end encrypted
- Zero-setup inference — no API key needed (fair-use). Processed in the clear, never stored
- Reflection reports — what changed about you, over time
What the cloud can and can't see
- Backup and agent relay: ciphertext.
- Encrypted on your machine with a key that never leaves it. We can't read them — not because we promise not to, but because we don't hold the key.
- Zero-setup inference: plaintext, in transit only.
- Running a model over your context means your text reaches our compute in the clear. There is no end-to-end encrypted way to do inference — that's arithmetic, not policy. So it's opt-in, we never store it, it never joins a training set, and the client that sends it is source-available so you can check what leaves.
- Don't want that? Then don't buy it.
- The free engine uses your own key, or a local model through Ollama, and never touches our servers at all. Everything that makes Persnally yours works without us.
The line we won't cross: the cloud may carry ciphertext and run jobs you asked for. It never becomes the place your context lives.
Privacy is never the paid tier. The local engine, the dashboard, and deletion stay free — Pro is convenience on top, not a wall around your own data.
Install, and see yourself.
One command finds your exports, reads your repos, synthesizes a profile, connects your AI clients, and opens the dashboard.
npm i -g persnally && persnally setupPaste in your terminal.
macOS · Linux · Windows · Node 20+ · background autostart on macOS & Linux · bring your own key, or run fully local with Ollama