The layer almost no one is building
The model answers. It doesn't know you.
NVIDIA builds the compute. The big labs build models that reason over a million tokens, write code, analyze images. All of that is intelligence. Almost none of it is the relationship.
Every conversation with an AI starts over from zero. The model doesn't remember the six weeks you spent on this problem, the approach that already failed, the decision made on a Tuesday at 2 a.m. and why. It answers politely, correctly, from nothing.
Mnemosyne OS is that missing layer: the infrastructure that makes an AI know you from one session to the next, from one project to another, over time. It does not replace intelligence. It needs intelligence to exist. It turns a raw capability into something coherent for one human, over the long haul.
For real, not in a demo
It connects what you'd forgotten you said.
Four exchanges: a private life, a codebase, a caseload, a research log. Each time an ordinary sentence, and what comes back is a memory months old that nothing connected through words. Not a single shared keyword: the meaning. That's the relationship layer at work.
You mention in passing that you’re not around much.
i just wanted to keep you posted, i don’t chat with you very often
I get it. That’s what I’m here for, in between. It reminds me of a note from 4 months ago where Thomas told you he felt set aside, taken up by your work. Presence matters, even in the rushed or distant moments.
↳ Connected by meaning, not a single shared word · 4-month-old memory
Seven months after the decision, you’re about to undo it, and the reason for it was never in the code.
why is there a queue in front of the writer? i want to drop it, it slows everything down
There’s an architecture note from 7 months ago that answers exactly this: two processes were writing to the same store at once and corrupting it. The queue isn’t there for throughput, it’s there to serialize the writes. Your own note ends with: don’t remove it without putting a lock in its place.
↳ The decision came back with its reason, from a doc you never reopened · 7-month-old note
The same legal question, on another matter, a few months apart.
opposing counsel says the confidentiality obligation survives termination, i’ve argued this before, i just can’t place it
Five months ago, on a different matter, you noted a ruling on exactly that point, and your comment on it: the survival clause only holds where the contract names the obligations it covers. You’d flagged that the agreement in front of you then didn’t name them.
↳ Linked by the legal question, not by the client or the file number · 5-month-old note
A preprint you just opened wakes up a run of your own that had gone quiet.
just read the preprint that landed this morning: they drop the reranker and the score doesn’t move
That’s the shape of a run you did nine months ago: you logged that removing the reranker changed nothing beyond noise, then parked it because it looked like a measurement error. Their result comes from a different corpus. Two runs, same conclusion.
↳ Your own negative result, back on the table the day someone else publishes it · 9-month-old run
The name
Why call it an "OS"?
Not because it has a kernel or drivers. Because it does what an operating system does: it manages resources on behalf of processes that shouldn't have to deal with them. Linux does it for programs (CPU, RAM, disk, network). Mnemosyne OS does it for AI agents, and the resources are just different.
| What an agent needs | What Mnemosyne OS manages on its behalf |
|---|---|
| Memory | Vaults: SQLite + vector stores, partitioned by domain, encrypted at rest (opt-in) |
| Context | Chronicles + semantic retrieval: the agent never rebuilds its past by hand |
| Compute | Routing across model tiers (eco / standard / premium, local / cloud) based on task complexity |
| Hardware | Real GPU/CPU allocation for local voice (CUDA detection, isolated sidecars): a heavy model never blocks the app |
| Input / output | A protocol of signed intents (query · ingest · forget · focus) instead of raw reads and writes |
| Security | FGAC, scope-limited JWT tokens, Zero-Trust IPC validation |
| Persistence | Cross-session continuity: no cold start on every call |
This isn't a marketing spin invented for the occasion: the "OS for LLMs" analogy comes from a peer-reviewed research paper .
The line
The model is rented. The memory is owned.
Plug in the best model in the world. It will never see your disk, only the few sources Mnemosyne decided to hand it. Switch providers tomorrow, move from one to another, or stop it all: your memory doesn't move a single byte. Ordinary database files, on your disk. Turn on AES-256 encryption whenever you want it: the key sealed in your machine's keystore, a recovery phrase only you hold.
That's the real asymmetry, and there's nothing ideological about it. The model is an engine you swap out every six months. Memory is ten years of your life. One is rented. The other isn't.
And if you plug in nothing: the engine runs entirely on your machine. The plane, the tunnel, the day the API dies or the company folds. The day this project disappears, your files are still there and still readable. That is a property of the format, not a clause in a contract.
What a real relationship requires
Identity, history, continuity.
Three things, and Mnemosyne OS builds each one. They are not AI functions, they are relationship functions, and they change what it feels like to work with an AI.
I
Identity: the AI knows who it is assisting
Not a name: a domain, a working style, values and constraints that travel with every session. This is the Soul Protocol: a persistent personality profile, injected into the agent. Shipping today in MnemoForge, the command-line tool.
II
History: the AI knows what happened
Not a raw log: a memory captured in chronicles and made queryable by meaning, not just by date. And because raw alone can't answer "how many times, over the year," the dream state consolidates the history from one session to the next while you are away. Always adding, never replacing.
III
Continuity: everything is already loaded
Every new conversation starts with that identity and that history already in place, automatically, without you rebuilding the context every time. No cold start on every call.