Hugging Face Announces Funes: Persistent Memory for Coding Agents

Hugging Face has announced the release of Funes, a new memory layer designed to enhance coding agents by enabling them to retain context and reasoning across sessions and machines. According to the announcement, Funes is a durable memory layer for agents like Claude Code, Codex, pi, and Hermes. It is built from the sessions already on your machine and works locally, becoming part of the agent's normal workflow with a single command.

The company states that Funes addresses the issue of agents losing context and reasoning when switching between sessions or machines. By indexing, retrieving, and ranking session logs, Funes allows agents to use this information while they work, providing a more reliable and efficient coding experience.

To use Funes, users can install it via the following command:

curl -fsSL https://huggingface.co/buckets/huggingface/funes/resolve/install.sh | sh

Once installed, users can add Funes to their preferred coding agent with a simple command, such as:

funes add claude

Funes is built on open-source technology and is designed to be integrated seamlessly into existing workflows. The company emphasizes that Funes is a dataset, not a service, and can travel to a Hugging Face dataset owned by the user, private by default.

David Corvoysier, the author of the announcement, highlights the benefits of persistent memory for coding agents, noting that it can significantly improve the reliability of agentic workflows. Johan Doc, a commenter on the post, also expressed interest in the concept and its potential to enhance enterprise workflows.

For more information, visit the Hugging Face blog.

Source: huggingface