Loading
Loading
Cognition does more than remember. It turns raw model capability into disciplined, project-specific execution: what to build next, what to avoid, how to verify, and which hard-won judgment to apply before your agents act. The layer that gets sharper the longer you run it.
Built by brains from
How the brain compounds
Every session feeds one loop: capture, approve, retrieve, apply, report. The science underneath decides what is slipping and reinforces it before it is lost, so trusted judgment compounds instead of evaporating.
The proof loop
Nothing reaches the team brain without an explicit yes. After that, each use reports whether it helped, so confirmed judgment drifts slower and one-offs decay faster.
Why it sticks
The retention span is read off the structure of what your team knows, not fitted or hard-coded. Detail fades fast, the principle persists. Drag the reinforcements: each review lands exactly where recall would lapse.
See the proofreinforcements
placed by the operator where recall would lapse, not on a fixed timer.
30d, left alone
4%
30d, reinforced
35%
derived span
S = 7d · β = 0.8
detail: commands, versions, edge cases
R(t) = exp(−(t/S)^β) · S = 1/λ₁ · derived, not fitted
A judgment layer, not a memory store
You never memorize a command. Say what you want in plain English and Cognition maps it to the right tool, loads the judgment that fits, and runs it. Storage is table stakes; deciding what to trust and applying it before your agents act is the product.
cognition_anticipatecognition_performancecognition_resolve_conflictcognition_recent_sessionsCognition is a governance layer, not a dumping ground. A skill only becomes team knowledge after a human signs off.
// governed.skill if (!humanApproved(skill)) return draftOnly(skill)
Get started
No dashboard to babysit. One install wires your agent to the shared brain, then it works in the background.
Create a key at /start. Solo or team, scoped to your host and project.
One command wires the MCP config and local instructions for your agent.
The agent verifies its tools, then loads approved skills before it guesses.
# one command, any MCP agent
$ COGNITION_KEY=cog_me_xxx curl -fsSL
https://www.cognitionus.com/api/install/codex | sh
For individuals
Preferences, decisions, fixes, and project scope stay available the moment a new chat starts.
Generate scoped keyFor teams
Approve reusable workflows, keep author attribution, and review outcome receipts before trust compounds.
Start a team pilot