Hugging Face's funes: a local, self-owned memory layer built from agent session logs
ariG23498 · x · 2026-09-07
Hugging Face detailed funes, a durable memory layer you own for coding agents (Claude Code, Codex, pi, Hermes), solving cross-machine and cross-agent amnesia.
Core idea
- Agent traces already record what changed and why, but they're only potential memory: you can't grep ten thousand turns to recover why you moved off a streaming parser.
- funes indexes existing sessions locally, adding retrieval, ranking and exact provenance with one command.
Usage & features
- Single binary; default inference backend has no ML runtime dependency — embedding and reranking run locally.
- Install via curl script, then funes add claude (or codex, pi, hermes) builds the first index, wires in recall/get tools, and auto-indexes each completed turn.
- No compaction, cross-machine, mid-session agent switching; optional sync to a private-by-default Hugging Face dataset you own.
More from coding & agent
- Anthropic Releases Free 4-Hour AI Engineering Course Focused on Claude Coding Workflows — Aiden_Tech_Ai · 2026-09-07
- Conductor recommended as a multi-model coding tool for Claude, Codex, GLM and Kimi — ayushtweetshere · 2026-09-07
- Open-source pipeline makes fabricated citations structurally impossible, full walkthrough released — Waste_Public_2985 · 2026-09-07
- Blogger claims Codex Desktop's harness architecture is a generation ahead of Claude — dotey · 2026-09-07
- Devs are talking to their AI agents like coworkers — positivity, avatars and small talk — menhguin · 2026-09-07
- Sierra launches τ^τ-Bench: coding agents must build real customer-service agents, gaps vs experts are large — sierra-research · 2026-09-07