Kleos is a research lab. We do R&D on the layer between a person and a model — the part that chooses what goes into the context window, what survives between sessions, and how much work it takes to get an answer.
We publish what we learn, and we ship the products it produces.
01
Two agents can finish the same task and spend wildly different amounts getting there. We measure what an agent actually consumes — tokens, calls, latency, local compute — and treat that as a first-class result rather than a footnote.
02
A decision made in one tool is invisible to the next. We work on memory that persists across sessions and across harnesses, stays inspectable, and knows when old evidence has stopped applying.
03
The test is not a benchmark task, it is a working day. Software, spreadsheets, research, communication — done repeatedly, so the system gets better at your recurring work instead of starting over.
desktop agent · open source
A local-first desktop agent. Give it an outcome and it plans the work, moves through your files and connected tools, and stops for your approval before anything important leaves the building.
copilot.kleosresearch.xyz ↗local agent memory · CLI + MCP
One local memory across coding agents and frameworks. Use the CLI, stdio MCP server, or Python and TypeScript clients while the proprietary engine stays on your machine.
Developer docs ↗