Brains remembers everything — your inbox, calendar, docs, chats, decisions — and turns that memory into boards, mini-sites, and agents you and your team can run. Any LLM. Any surface. Bring your own model.
Most "AI agents" forget you every conversation. Brains doesn't. Every email, calendar event, doc, message, and prior chat becomes a queryable page your agents reason on — persisted, indexed, ready.
Connect a source — brains handles the rest. Ingested, structured, searchable, and yours alone.
Connect once with OAuth. Brains backfills history, then keeps it fresh via scheduled syncing + on-demand fetch. No manual upload, no re-pasting.
Each item becomes a queryable page with structured metadata — sender, date, attendees, source. Your brain is a knowledge graph, not a blob.
Find anything by keyword, by meaning, or by direct reference. Full-text search for exact phrases. Semantic search for concepts. Agents pick the right one.
Memory is scoped to your brain. Never trains anyone's model. Share specific brains with teammates via owner / contributor / observer roles.
The codex is brains' developer platform — a growing library of installable recipes for boards, mini-sites, automations, and workflows. Real TypeScript code, not no-code templates. Describe what you want; brains composes recipes into a working tool in seconds — with idempotency, versioning, and safety baked in. Spin up an agentic workflow as fast as you'd write a prompt.
Each recipe is sandboxed TypeScript with scoped MCP tokens. Idempotency and two-phase safety baked in — automations never silently send an email.
Recipes ship like packages — install in one click, upgrade like software, share with your team. Your team's tools compound.
Every prompt either composes existing recipes or expands what brains can build next time. The codex compounds with every team using it.
Track anything with typed rows, statuses, and relationships. The board is the database.
Visualize what matters with cards, charts, and kanban — over any board. Updates as data changes.
Published HTML you can share with a link. Investor updates, all-hands, customer pitches. One click to share.
Goal-shaped containers: charter, roster, KPIs, deadlines — plus the agents to drive them. Flip to active. Work flows.
Recurring summaries delivered on schedule. Inbox-triage, calendar-aware, two-phase-safe. Wake up to insight.
One-shot deep work — competitive scans, due-diligence dossiers, market analyses. Ground truth, cited sources.
Workflows package a goal, a roster, KPIs, deadlines, a board, and the agents to drive them as one shippable object. Pause, share, hand off.
Define what you're shipping, who's on it, what success looks like, and when it's due. Brains attaches the automations — progress scorer, milestone reviewer, task nudger — that keep it moving. Flip to active and the team runs.
Share a brain, a board, a mini-site, or a workflow. Owner / contributor / observer roles cascade automatically. Your team's agents and their data inherit the same permissions.
Boards are typed datasets that automations read and write. Mini-sites publish them as live dashboards your team — and the outside world — can actually use.
Scheduled TypeScript programs with scoped MCP tokens, idempotency baked in, and cost caps. They read your memory, write to your boards, and never silently send an email.
Your brain lives outside any single chat. Use it from Claude, ChatGPT, Cursor, Gemini, Telegram, or your terminal. Switch models, keep your context.
Open Claude on your laptop, ChatGPT on your phone, Telegram on the go, Cursor in your editor, the terminal in CI. Your brain follows you everywhere.
This is what "agentic" actually means: not a single chatbot, but a substrate your entire stack can reason from — through any LLM, on any surface.
MCP-native, model-agnostic, platform-portable. The work doesn't live in the model. It lives in brains.
Memory that remembers everything. Boards, mini-sites & workflows in one prompt. Any LLM, any surface.