HuggingFace released Funes, a local memory system for coding agents, on September 17, 2026, allowing agents to recall past sessions without relying on cloud-based models. It is the company's first major update to its coding agent tools since the launch of its initial agent framework in 2025.
Funes indexes traces from Claude Code, Codex, pi, and Hermes into a local Lance dataset, enabling agents to access previous reasoning without LLM summarization. The system allows agents to recall past sessions and use the data for future tasks, enhancing the continuity of coding workflows.
"Your coding agent solved this once. It found the failing test, tried the obvious fix, rejected it, learned the weird constraint, and made a decision," said Aritra Roy Gosthipaty, author of the article. The system ensures that agents can refer back to past decisions without losing the original context.
The announcement follows the release of the Lance community tools, which HuggingFace said are designed to support local and private data handling for developers. Funes is positioned as a privacy-first solution that keeps agent traces local by default, with optional sharing capabilities.
HuggingFace did not specify how the system will scale with large datasets, and it raised the question of how Funes will handle complex memory retrieval in real-time. The company said the system is designed to be a library rather than a standalone service, allowing developers to integrate it into their existing workflows.
Source: huggingface
