# brethof-brain

> A memory system for AI agents. Memory that is already there when a new session starts, that hands the agent what matters the moment it is relevant, and that keeps everything said so it can be found again. We only process your memory. We never store it: it lives on your own machine, or in our cloud encrypted under a passphrase only you hold. The processing runs in a secure compute instance in Zurich, Switzerland, and keeps nothing of it; the thinking is done by a model under zero data retention.

brethof-brain stores memory for anything an agent does — coding agents are the first audience, not the limit of it. One account powers every agent you run, on every machine. It is made by Brethof AI, a company in Poland, in the EU.

Why it is different:
- We only process your memory. We never store it. Local edition: the memory is stored on your own machine. Hosted edition: encrypted under a key derived from a passphrase only you hold; locked, it is unreadable to everyone, us included. Our hub reads a piece only when your box sends it one to work on, and keeps none of it — durably it holds only your standing rules and one line per project.
- The processing runs in a secure compute instance in Zurich, Switzerland (a Tier III+ datacenter). Every conversation passes through it — that is how memory gets made — but it only processes, holds each exchange only long enough to be understood, and keeps none of it. It stores only your standing rules and each project's one-line purpose.
- The thinking is done by a model whose provider is bound by contract not to log, retain or train on what passes through (stated as the provider's commitment).
- It is already there, most of the time: rules arrive with almost every message, the records that bear on a prompt arrive with it, a brief opens every session — and everything ever said can be searched.

What you get — the seven things:
- Curated records: the current truth of each project, handed to the agent with every message they bear on. The memory curates itself, reconciling new facts against old ones.
- Rules: standing instructions, loaded with most messages; every agent on every project works the same way. Saving or changing a rule costs one message.
- The full chat history: every session kept in full and searchable; protected — it cannot be deleted piecemeal, not even by your own agent.
- The graph — the memory of decisions (included in every plan): every decision dated, with its reason and what it replaced, and what was tried, rejected or dropped, so nobody re-proposes it. Records keep the truth of now; the graph keeps how it got there. It reaches the agent inside search, as up to five dated history lines under the records. History is never edited: a reversal is a new line.
- Notes: the agent's own handover, written by it, loaded at the start of the next session ("say continue").
- Playbooks: a procedure written once and run by any agent on any project; never pruned. The worked example is the atlas — a map of your whole system kept as playbooks (see /brain/playbooks/). Three methods ship with every memory, on both editions: atlas (the map), plan (what is decided and next) and prospectus (a product's sheet).
- The Ledger: catches, in the background, the things that were said and meant, and keeps itself clean of repeats and of rows the later work shows done; you check it later; a row that still matters moves into the project's plan, the rest is deleted.
- Around them: memory that arrives on its own at session start (rules, projects, last notes, an index of what is known); organised by project with a one-line purpose (starting a project costs ten messages); scoped keys (a key limited to chosen projects, read-only or read-write); you own the data (export, delete a project, redact an exact phrase, delete the account — from the panel); it degrades softly (if our hub is unreachable, the memory keeps recording and your agent's own search keeps answering; the session brief and per-prompt memory wait for the hub).

Editions — one choice at signup, not switchable later (plans and packs can change any time):
- Local: everything — the memory, the archive, the graph, the search — is stored on your machine. Needs Docker or Podman, about 3 GB of disk including the memory model (downloaded once), a passphrase and a key from the panel; no elevated privileges. Storage is your own disk, unlimited by us.
- Hosted: the whole memory is encrypted under a key derived from your passphrase; it unlocks when you start a session and idle-locks after a window you choose. Stored and searched in Germany; backups are the encrypted memory itself, in Germany, deleted when the account is. If the passphrase is forgotten, no one can open it — deleting the account still works.

Install: sign up at brethof.ai/account (email and password, or GitHub or Google), choose local or hosted, set your passphrase, copy your key — then tell your agent "install brethof-brain"; a small window on your own screen takes the key (and, hosted, the passphrase), so the agent never sees either. Twenty proven platforms (green on a clean machine with the newest release of each, installed from GitHub and npm). On Linux and Windows: Claude Code, Codex, Qwen Code, OpenCode (1.x and 2.0), Kilo Code, DeepSeek Harness (dsh), OpenClaw, Hermes Agent (a Hermes memory provider), GitHub Copilot CLI, Pi, CodeBuddy, Devin CLI, Kiro CLI, Qoder CLI, goose, Gemini CLI (a Gemini CLI extension) and Antigravity CLI. On Linux: Cline, Amp and MiniMax Code. The client is v1.2.20, source-available (read every line), installed as a plugin or from the GitHub repository (pip, pinnable by tag @v1.2.20; no PyPI package). The Claude Code plugin ships three commands: /recall, /onboard, /curate, and never edits Claude Code's own files: the memory block for ~/.claude/CLAUDE.md is a recommended manual step from its README. The pip install (brethof-brain install-hooks) keeps that block itself, between its own markers; with it, "remember this" goes to the Brain, not to Claude Code's own memory files.

Plans and prices (USD per month, the same everywhere on Earth and on both editions):
- The unit is the message — one exchange, a prompt and its reply; no token counts. One message covers up to 12,000 characters. The graph of decisions is included in every plan; one list for both editions.
- Free $0: 500 messages a month (local and hosted), 20 MB hosted storage.
- Starter — full price $14.99, founder price $9.99: 2,000 messages, 500 MB hosted storage.
- Pro — full price $29.99, founder price $19.99: 4,000 messages, 2 GB.
- Team — full price $74.99, founder price $49.99: 10,000 messages, 5 GB.
- Business — full price $149.99, founder price $99.99: 20,000 messages, 15 GB.
- Founder prices are the discount live now; anyone who subscribes at a founder price keeps it for as long as they stay subscribed. Packs (one-time, never expire): 2,000 messages for $9.99, 5,000 for $24.99. Hosted storage top-up: +1 GB for $5.99 a month.
- At the limit the learning stops (no new records, rules, ledger rows); the session brief, the rules and records on every prompt, search, the full history, the graph, notes and playbooks keep working. No hard wall, no overage charge. Idle free accounts and lapsed subscriptions: archived after 7 days, warned around day 23, removed on day 30.

What we do not claim: the local edition still runs a container on your machine; conversations do pass through the hub for processing (what never reaches us readable is the memory); zero data retention is the model providers' stated commitment (Ollama for the model, Cloudflare's catalog for TypeSafe's Jev), not our guarantee; the client is source-available, not open source; a platform is listed only when proven on a real install.

## Pages

- [brethof-brain](https://brethof.ai/brain/): The product — why it is different, the seven things, the two editions, privacy and security
- [How it works](https://brethof.ai/brain/how-it-works/): How memory arrives on its own, curates itself, and where the processing runs
- [Projects & rules](https://brethof.ai/brain/projects-rules/): Organising by project, teaching with rules, notes, scoped keys
- [Playbooks](https://brethof.ai/brain/playbooks/): Procedures written once and run everywhere; the atlas as the worked example; how to start in one sentence
- [Install](https://brethof.ai/brain/install/): Get a key, choose an edition, tell your agent to install it; proven platforms and local requirements
- [Pricing](https://brethof.ai/brain/pricing/): Plans, allowances per edition, packs, what happens at the limit, full and founder prices
- [Docs](https://brethof.ai/brain/docs/): Quickstart per platform, keys and projects, day-to-day use, the Ledger, playbooks, rules, space and usage, FAQ
- [Privacy Policy](https://brethof.ai/brain/privacy/): What is processed where, sub-processors, retention, your rights
- [Terms of Service](https://brethof.ai/brain/terms/): The agreement, plans and payments, the archive protections
- [Security](https://brethof.ai/brain/security/): Encryption, the passphrase, the hub, what leaves your machine and what never does

## Company

- [Brethof AI](https://brethof.ai/): Privacy-first AI from Poland — Brethof Voice Pro, brethof-brain, guides, Nova
- [Account panel](https://brethof.ai/account/): Sign up, choose your edition, create keys, export or delete your data
- [Client repository](https://github.com/BrethofAI/brethof-brain-client): The source-available client, adapters for every proven platform