Selected work
What we build
We judge the work by what runs. These are systems we have designed and run in production ourselves. Named client engagements are added as they ship.
01
Multi-agent operations fleet
Around twenty specialised agents running unattended across triage, code, review, research, and operations. Shared memory, self-dispatch, and run history visible from one control plane.
02
Long-term agent memory
A retrieval system over thousands of prior decisions, ranked by relevance, recency, and outcome, so agents stop repeating old mistakes.
03
Automation pipelines
Scheduled and event-driven jobs that fetch, transform, and act, with health checks that page a human only when they should.
04
Production guardrails
Evaluation harnesses, cost controls, and policy gates that sit between a model and anything it can change.