What is funes used for?

funes is used to create a searchable memory layer for AI coding agents. It indexes supported agent session traces so agents can recall relevant prior work with links back to the original source turns.

Which AI coding agents does funes support?

The provided information states that funes indexes session traces from Claude Code, Codex, pi, and Hermes. Support may depend on the trace format and the current state of the open-source project.

Yes, the supplied information lists funes as free and open source. Users should check the GitHub repository for the latest license, installation requirements, and any related service costs such as optional hosting or storage.

How does funes retrieve memories?

funes combines local vector retrieval with BM25 keyword search. It also includes reranking, recency weighting, and provenance-linked source turns to help agents retrieve useful and traceable context.

Can funes work across multiple machines?

The supplied information says funes can optionally sync a redacted memory dataset to a private-by-default Hugging Face Hub repository. This is intended to support cross-machine and team recall, but teams should verify configuration and privacy behavior before using it with sensitive data.

Last updated: 9/22/2026Last verified: 2026-09-04

funes Review

8views

0saved

Open-source, local-first memory layer for AI coding agents that indexes Claude Code, Codex, pi, and Hermes session traces into a searchable Lance dataset, giving agents provenance-linked recall across sessions and machines.

Free

What is funes?

funes is an open-source, local-first memory layer for AI coding agents. It indexes session traces from supported tools such as Claude Code, Codex, pi, and Hermes into a searchable Lance dataset. The goal is to give coding agents provenance-linked recall across sessions and, optionally, across machines or teams.

Key Features

Indexes supported coding-agent session traces into a durable, searchable local memory
Uses a local Lance dataset to store and search indexed coding-agent history
Combines vector search and BM25 retrieval for hybrid recall
Supports reranking, recency weighting, and provenance-linked source turns
Adds recall and get tools to supported agent workflows
Supports incremental per-turn indexing so memory can update as sessions progress
Can optionally sync a redacted memory dataset to a private-by-default Hugging Face Hub repository for cross-machine or team recall

Best For

Developers who use multiple AI coding agents and want shared recall across sessionsTeams experimenting with local-first memory for coding-agent workflowsOpen-source users who want inspectable infrastructure for AI coding memoryDevelopers who need provenance-linked references back to prior agent conversationsUsers who want local retrieval with optional cross-machine synchronization

Pricing

funes is listed as a free, open-source project on GitHub. No paid tiers or commercial pricing details are provided in the supplied information.

Pros & Cons

Pros

Open-source and local-first, making it suitable for developers who prefer inspectable tooling
Focuses specifically on memory for AI coding agents rather than general note storage
Hybrid vector and BM25 retrieval may help surface both semantic and keyword-based matches
Provenance-linked source turns make recalled information easier to trace back to prior sessions
Optional Hugging Face Hub sync can support cross-machine and team recall workflows

Cons

Support appears limited to specific coding-agent session trace formats such as Claude Code, Codex, pi, and Hermes
As a GitHub-based open-source project, setup and maintenance may require developer comfort with local tooling
The supplied information does not describe a hosted user interface or managed service
Teams may need to review redaction, syncing, and repository permissions before using shared memory workflows
Effectiveness may depend on the quality and completeness of indexed agent session traces

Alternatives

Continue

Continue is an open-source AI coding assistant that integrates with IDEs and can be configured with different models, making it relevant for developers building customizable coding workflows.

Cursor

Cursor is an AI-focused code editor with codebase-aware assistance, chat, and editing features for developers who want an integrated AI coding environment.

GitHub Copilot

GitHub Copilot provides AI code completion and chat inside popular development environments, making it a mainstream alternative for AI-assisted programming.

Sourcegraph Cody

Cody is an AI coding assistant designed to work with codebase context, which may appeal to teams that need code search and AI assistance together.

Pieces for Developers

Pieces offers developer productivity tools with saved snippets, workflow context, and AI assistance, making it relevant for users interested in persistent coding context.

FAQ

View all funes alternatives →

Details

Platform

macOSLinux

Features

Indexes supported coding-agent session traces into one durable, searchable local memory
Combines local vector and BM25 retrieval with reranking, recency weighting, and provenance-linked source turns
Adds recall and get tools plus incremental per-turn indexing to supported agent workflows
Optionally syncs a redacted memory dataset to a private-by-default Hugging Face Hub repository for cross-machine and team recall

Languages

Known limitations

It currently documents direct integrations for Claude Code, Codex, pi, and Hermes; other agents need a compatible trace export or MCP-style integration
First indexing and local embedding/reranking-model downloads can take time and consume local disk space
The release installer publishes prebuilt binaries for Linux x86_64, Linux aarch64, and macOS Apple Silicon, so other platforms may need a source build
Hub syncing is optional but requires a Hugging Face dataset repository and its credential/access configuration