JCOM Labs is an independent open-source lab building local-first AI systems, developer tools, and practical improvements to efficient LLM inference.
We care about systems that people can inspect, reproduce, and control. That means clear trust boundaries, deterministic tests around AI behavior, evidence-backed performance claims, and honest documentation of limitations.
- Axiomux — run one agent across providers, skills, governed memory, fusion strategies, and safe computer-use boundaries.
- Window Swap — remove a small but persistent friction point from snapped-window workflows on Windows.
- Code Hangar — turn scattered AI coding sessions into safe, reusable local project knowledge, with evidence-first review and explicit privacy boundaries.
- MoE Autopilot Studio — a Windows-first evidence lab for measured MoE placement decisions, published as the reproducible companion to the broader MoE Autopilot research line.
The core MoE Autopilot research repository is being prepared for public release.
JCOM Labs is maintained by JC-OM in Portugal.