A local-first memory stack for AI agents — and a way to see it work.
Two pieces, one idea: give an agent a memory you can actually trust and inspect, then watch it operate in real time.
| Repo | What it is |
|---|---|
| markmem | The engine. A self-learning AI memory on Markdown + git: HybridRAG retrieval, a real learning loop, and nothing enters memory without human review. A memory an auditor can read. |
| orbital | The view. A real-time 3D mission-control for an agent operation — sessions, agents, pipelines, tools, memory and infrastructure as one live particle graph. |
Why together. markmem is the brain; orbital is the window into it. The engine makes the memory trustworthy and local; the viewer makes it legible — you see which memories fire, when, and why. Each stands alone; together they turn "the agent remembered something" into something you can audit and watch.
Built to grow. orbital already renders sessions, agents, pipelines, tools and infrastructure as first-class nodes — so each new piece of the operation joins the same live picture as it lands.
Shared principles. Local-first — your data never leaves your machine. Zero runtime dependencies (stdlib core). MIT. Privacy as structure, not a promise.