The whole design rests on one idea: you stop being your own memory.
Every agent session is a conversation flowing in. The hooks turn that into a circulation. Everything that comes in is archived; everything the memory knows is brought back out — without you asking on either side. You never see the plumbing, you just notice it's there when you don't have to re-explain yourself.
Your messages, your agent's replies, its thinking, and a one-line marker saying which tool it used — all of it becomes part of the transcript. The contents of the files it read and the output of the commands it ran are dropped before anything is sent. If you want proof, the client is a few hundred lines and you can read every one of them. The Brain does not summarize. It keeps the complete transcript of every session and makes it searchable, so recall never depends on what survived a summary or fit in a context window.
Every session starts with your agent already briefed on the project — its rules, where things stand, what was decided.
And before every prompt, the matching memory is pulled in so the agent has it when it starts answering.
No tool calls, no hunting, no remembering to ask.
Never explain your project twice — not to any agent, not on any machine. Start the PC, open the session, say "continue", and you're back where you left off. Your agent knows the rules, the decisions, the current state. Only you are allowed to forget.
Every prompt pulls the matching memory in before the agent starts answering. It spends no tool calls, no turns, no hunting — the memory is simply there. Every other memory store waits to be asked, and a store your agent has to query is only as good as the model's willingness to query it. This one does the work itself, whether or not anyone remembers to ask.
After every exchange, a curator reads the turn and updates the memory: decisions saved, changed facts folded in, claims reality has overtaken deleted. Seconds, not sessions. Nothing has to be filed by hand for the memory to build itself.
When you *want* something remembered on the spot, you say so and your agent records it. But that's a convenience, never a requirement.
Your memory lives in two places, and the pairing is the point. Curated memory is the current truth — dense, up to date, the thing your agent works from. The conversation history is the complete raw recording — everything that was ever said. One is fast but lossy; the other is the insurance.
This is where the agent finds what's true right now. When you reverse a decision, it's reflected here — and in the graph, the old path is flagged as a dead end. An agent steering toward it is stopped mid-prompt. Your agent stops proposing the thing you killed in March.
Every session, kept in full. No compression, no summary that decides what matters. When you need to know exactly what was said — or why a decision was made — it's all there, searchable by keyword, by meaning, or through the graph.
You never have to write anything down. The curated memory stays current on its own, and the recording is always there underneath. A note is just a convenience on top — and if you forget to take one, the conversation that produced the moment is already in the archive.
Exact words. When you remember the phrase, find the phrase. Exact-match search across the whole memory.
When you only remember what it was *about*, semantic search finds it anyway. You don't need to get the wording right.
The graph maps the files, services, decisions and errors your sessions have touched. Use it when you're looking for something you tried before, or for the pattern running across a project — what led to what, what was killed, what came of it.
The Ledger watches your conversations and catches the things you said you'd do and never did. The gap between talking and doing is where work dies. This is the thing that makes sure it doesn't.
As conversations are archived, the Ledger watches for intentions — "I'll fix that later," "we should follow up on this" — and each becomes a row waiting for you. Nothing you say out loud just disappears.
Say “check the ledger” and your agent walks the rows with a verdict already attached to each — done (with the evidence), pointless now, still worth doing. You answer in single words. Fifteen rows takes about two minutes. Every decision removes its row, so the pile only ever shrinks. No statuses, no stale board.
The rows you keep become goal briefs your agent runs to completion on its own. Review the ledger in the evening; the work happens while you sleep.
No statuses, no stale task board, no pile — the ledger holds only what’s undecided, and every decision shrinks it. And nothing is ever locked out: delete an idea today, and if it comes up again next month — because the project moved and now it fits — the Brain files it again, judged fresh.
The deploy, the weekly report, the translation run — write it down once as a markdown playbook, stored for your whole account.
Any of your agents, in any project, on any machine, can fetch it and run it. Projects are separate; procedures are shared.
The Brain never runs the playbook itself — there's no execution backend, no terminal, no access to your machines. Your agent does the work.
Playbooks are yours entirely — four doors, nothing more: save (create and edit — saving under the same name replaces the previous version), get, list, delete.
Once a week, the healer reads your whole memory at once — the one thing a curator watching a single exchange can never do —
and merges records that forked apart,
absorbs duplicates written months apart,
and reconciles the places where memory disagrees with itself.
Your memory heals itself, on its own.
Nothing to run, nothing to remember.
You don’t manage any of it. That is the product.
Local speech-to-text that learns your voice. Perpetual licence. Our flagship.
PAID · flagship
Long-term memory for your AI agents — full-text + vector + graph. Briefed at session start, recalled on every prompt, archived automatically.
PAID · free tier
Print-ready digital models. GLB and OBJ included. Lifetime access.
PAID · digital catalog
Five printers and real capacity. No self-serve shop yet — tell us what you need and we quote it by email.
BY ARRANGEMENT · email us
Our YouTube channel. A cyber-tiger host walks through local AI tools and what they actually do.
CHANNEL · live
Curated GitHub lists for AI coding agents, MCP servers, local AI and Linux for AI. Every entry carries a source link.
FREE · curated
Long-form how-tos for local AI on Linux, Windows and macOS, with the configuration files included.
FREE · live
Negative-curation: practices and tools that waste your time, ranked. Receipts required.
FREE · live
Who we are, why we build privacy-first AI, and what we won't do.