Tools I've shipped recently.
evav
An AI research lab studying agents that actually remember.
Independent lab focused on long-horizon agents, verifiable tool use, and the unglamorous infrastructure of memory.
evav/mcp
Model Context Protocol server for evav tooling.
Expose the lab's research utilities — retrieval, memory snapshots, evaluation harnesses — to any MCP-compatible agent. Drop-in, well-typed, well-documented.
evav/benchmark
A public leaderboard for agentic systems.
Live leaderboard tracking agentic systems across the evav eval suite. Reproducible runs, transparent scoring, no marketing fluff.
atlas-rs
A small, fast vector store in Rust.
Single-file embeddable vector index. No daemon, no kubernetes, no excuses.
marginalia
Reading-as-thinking, for papers.
A minimalist PDF annotator that treats your marginal notes as first-class objects.
evalboard
Evaluation pipelines that you can read.
Composable eval framework that prizes legibility over magic. Used internally at evav.