Pylit builds agentic research infrastructure for scientific discovery.
We build systems that help humans and AI agents preserve context, test ideas, learn from failures, coordinate work, and compound execution evidence into reusable logic.
Our goal is to make scientific and technical work more cumulative: more experiments, better feedback loops, faster learning, and more durable progress across days, months, and years.
We build in the open, test in real workflows, and keep what compounds.
- Research memory systems
- Agent coordination systems
- Evaluation and experiment loops
- Scientific discovery workflows
- Reusable reasoning infrastructure
- Human-AI collaboration tools
- OpenDream — persistent memory for research systems: traces, reviews, and graph-backed context that survive across sessions.
- Model Preflight — evidence gates for models before deployment: probes, capability checks, and go/hold reports for serious workflows.
- metactl — control plane for reusable agent behavior: skills, rules, workflows, packs, and audits that turn lessons into machinery.
AI should expand the frontier of what humans can discover and compress not only knowledge, but time.