Notes on building the engine.
Research, architecture, and field notes from the people designing and operating Lumeenn — published as we ship, not after the fact.
Inside the policy layer that gates every agent action
How Lumeenn checks spend limits, data boundaries, and approvals before any irreversible step executes.
Read articleWhat changed when we let the model see its own plan
A small architectural change to plan visibility cut task failure rates significantly. Here's what we learned.
Read articleLumeenn raises Series C to scale the reasoning engine
New capital goes toward dedicated capacity, expanded connectors, and doubling our applied research team.
Read articleBenchmarking judgment: why accuracy alone misleads
A task can be "correct" and still be the wrong call. We propose a way to score decisions, not just outputs.
Read articleCutting median reasoning latency from 600ms to 140ms
A breakdown of the caching, speculative planning, and routing changes behind our latest performance jump.
Read articleDesigning an undo button for autonomous agents
Reversibility turned out to be the single biggest lever for user trust. Here's how we built it in.
Read article