ThinkingRoot DB
The cognition database — a managed cloud for AI products that answer from your real data and admit what they don't know.
ThinkingRoot DB is a cognition database: you feed it sources, it compiles them into a graph of atomic, source-locked claims, and your AI reads from that graph instead of guessing. Answers cite the claims they came from, and when the graph doesn't contain an answer, the system says so instead of fabricating one.
ThinkingRoot DB Cloud is the managed, multi-tenant home for that engine. It gives every project its own isolated engine instance, an authenticating API gateway, a secrets vault, a connector catalog, and this Console — so you can build on the cognition graph without operating it yourself.
What this is, in one sentence
A database whose unit of storage is a grounded claim, exposed to your AI over REST and MCP, hosted so you don't have to run it.
Pick your path
Get Started
Create a project, compile your first source, and ask a grounded question in about five minutes.
Core Concepts
Sources, claims, witnesses, entities, branches — the mental model that makes the rest click.
Connect your AI tool
Copy-paste MCP configs for Claude Code, Claude Desktop, Cursor, and Codex.
API Reference
Authentication, the response envelope, and the engine REST surface.
What you can build on it
- Assistants that answer from your codebase, docs, or tickets — and cite the exact claim behind every answer, so you can trust and audit them.
- Agents that remember and self-extend — each working session gets its own branch that auto-merges back, and agents can install MCP servers, define skills, and deploy Root Functions to acquire new capabilities at runtime.
- Products that won't fabricate — the engine returns empty state, never invented rows. A 404 is an empty list, not a lie.
How the pieces fit
| Layer | What it does | Where it runs |
|---|---|---|
| Console | The UI you're reading docs in — projects, playground, keys, connectors, billing | console (Next.js) |
| Gateway | One authenticated entry point; routes tr_sk_ keys and session JWTs to the right place | cloud control plane |
| Provisioner | Spawns one engine container per project, injects secrets, enforces the egress allowlist | cloud control plane |
| Engine | The cognition database itself — compile, query, branch, MCP. Open source (root) | one isolated container per project |
The engine is open source and self-hostable (root serve); the cloud adds the
multi-tenant vault, gateway, UI, and billing on top. See
Cloud vs Self-host for the exact split.