Funes is an open-source memory layer for AI coding agents, released by Hugging Face on GitHub. Named after Borges’ character with perfect memory, it indexes your past agent sessions and makes them searchable, so any agent can recall the decisions, rationale and findings from earlier work. The project has around 570 stars and is under very active development in 2026.
Coding agents forget everything between sessions, so teams keep re-litigating the same decisions. Funes fixes that by ingesting session transcripts from Claude Code, Codex, pi and Hermes into a single memory store built on Lance, then exposing recall to agents — typically over MCP — so the model can spontaneously pull in what was decided last week and why. Memory is treated as a dataset: you can publish it to the Hugging Face Hub, after which any machine, teammate or agent can recall from it, and answers come back grounded in the actual sessions they drew from. It is written in Rust for fast indexing and search, and is append-only by design, preserving the full history rather than summaries alone.
The name nods to Funes the Memorious, the Borges character who forgets nothing — and that is the design goal: nothing an agent learned is lost when a session ends. Because memories are ordinary datasets, they can be versioned, browsed and shared like any other Hugging Face dataset, turning one developer’s agent history into a reusable team asset.
Funes is free and open source under the Apache-2.0 license. Publishing memories uses your own Hugging Face Hub account under its normal terms.
Developers and teams who work with coding agents daily and want those agents to remember project history — past choices and their reasoning — across tools and machines.