Open infrastructure for AI‑agent memory control.
An aquifer is where water is stored under pressure and rises without pumping when a well reaches it — our metaphor for agent context: the right knowledge surfaces automatically, without flooding the prompt.
- artesian — memory control for AI agents. A Rust, MCP‑first system that layers a control plane over any retrieval store: bounded committed context, qualify‑gate, pluggable backends (files / sqlite‑vec / Qdrant), hybrid retrieval (~1 k tokens/query), optional master·worker·judge orchestration (wellfield teams), autonomous run‑until‑done loops, and self‑repair across context compaction. Apache‑2.0. 📖 Documentation »
- ocf — Open Cognitive Format. A portable snapshot of what an agent holds in force right now — bounded committed working state + a qualify log — so any runtime can resume rather than re‑retrieve. Composes with PAM / AMP / files; artesian is the reference implementation. CC‑BY‑4.0 / Apache‑2.0.
- artesian-zed — Zed extension that registers Artesian's MCP server as a Zed context server, giving Zed's agent durable, qualify‑gated memory.
- homebrew-tap — install Artesian with one line.
brew install aquifer-labs/tap/artesianThen follow the documentation to wire it into your agent.
memory control · non‑intrusive · MCP‑first · Rust · pluggable backends · open (Apache‑2.0)
Use as little or as much as you want: just retrieval, or the full agent loop with a qualify gate.