Quickstart
From a new account to a grounded, cited answer in about five minutes.
This walks through the Console. Everything here is also available over the
REST API and the root CLI if you prefer.
Create an organization
Sign up and create an organization. Orgs own projects, billing, and team membership. You can rename it later in Settings.
Create a project
Give the project a name. Behind the scenes the provisioner spawns a dedicated engine container for it and the Console polls until it reports running. When it goes green, your cognition database is live.
If the status sits on provisioning for a while, the engine image is being pulled or booted. The Overview tab shows the live engine status and version.
Add a source
Open the Sources tab and add a source — a file, a document, or text. A source is the raw material the engine compiles into claims. Nothing is extracted yet; you've only registered where the knowledge comes from.
Compile
Trigger a compile. The engine reads the source, extracts atomic claims (each locked to the exact span it came from), links entities and relations, and runs the result through its admission checks. When it finishes, the Overview cards show non-zero claim and entity counts.
See Sources & compilation for what happens inside a compile.
Ask a grounded question
Open the Playground and ask something your source actually covers. The answer comes back with witness chips — the claims it was built from. Ask something your source does not cover and the engine will tell you it doesn't know, rather than guess.
(Optional) Connect your AI tool
In MCP & Agents, copy the config for your client (Claude Code, Cursor, Codex…). Now your assistant reads from the same grounded graph. See Connect your AI tool.
What you just built
You now have a project whose engine holds a graph of source-locked claims, an API key the Console is already using, and an MCP endpoint any compatible AI tool can connect to. From here: