Skip to main content
SenseLab makes agents improve recursively: the output of every run — what was learned, what was decided, and how it turned out — becomes the input to the next one. Nothing in this section is “a database feature”. Each concept is one stage of that loop.

​
The loop

The loop: Brief, Act, Record, Outcome, Update — and on every run the next brief starts from what this one proved
The fast loop closes on every run. An agent is briefed on what the account already knows, works, and reports how it went. Confidence moves on everything it read, so the next agent to touch the same subject is ranked into what has worked and away from what has not — without anyone editing a prompt. Two slower loops run off the same record. Behavior grades runs with judges, opens incidents on recurring failures, and proposes fixes to the learning itself — tested by replay, shipped by merge, rolled back if they regress. Managed Models turn the decisions that worked into a fine-tuned model of your own, which then does the work and feeds the next tune. Each loop improves the input to the others; that is what makes the improvement recursive rather than one-off.

​
Who is who

​
What an entry looks like

Anatomy of a learning entry: entity_path, key, value, memory_type, version, confidence, evidence_status, branch and provenance, with the write event it logs on the agent's timeline
Learning is private to your account by default. Rooms are how you pool it, so several people’s agents — on Claude, ChatGPT, Cursor or your own code — run the loop over the same knowledge.

Rooms

Shared workspaces where several people’s agents learn from the same entries.

Behavior

Judges, incidents and fixes — the loop that repairs the learning itself.

Instrument your agent

Get every run into the record with steps, LLM calls and cost.

Managed Models

A private model trained on the decisions that worked.