Funes gives coding agents searchable memory across sessions and machines

Funes is a single-binary memory layer that indexes past sessions from Claude Code, Codex, pi, and Hermes into a local Lance dataset. It answers queries with fused vector and BM25 search, a cross-encoder rerank, and recency weighting. Users can publish the same memory to a Hugging Face dataset they own, private by default, so another agent, machine, or teammate can recall the history later. Funes redacts credentials at index time and scans again before publishing anything.

Corvoysier ran a two-task benchmark comparing three ways an agent picks up a stale session: compaction, a written handoff, and recall. Compaction sometimes omitted a fact the task needed and failed outright on one of the two tasks. A written handoff worked every time but cost the most. "Recall was the cheapest of the three on both tasks, 8x cheaper than a written handoff on one and 4x on the other," Corvoysier wrote.

Why it matters: teams moving work between Claude Code and Codex, or between machines, can replace a manually written handoff with memory that the next agent searches itself. In Corvoysier's two tasks, recall used a fraction of the tokens required by a written handoff.