Hugging Face has released 'Funes,' an open-source tool that helps coding AI agents permanently store and reuse their past decisions and work context in a local environment.
Imagine this: yesterday, you designed a complex payment system for a website with your AI coding agent. But this morning, you have to explain the entire task from scratch to an AI that doesn’t remember the work. Just like meeting a new person every morning, your valuable work time is often wasted because of the AI’s ‘forgetfulness.’
Recently, Hugging Face, a hub for the AI community, released an interesting tool to solve this exact problem. It is called ‘Funes.’ Give Your Coding Agents a Memory You Own - Hugging Face Funes is a ‘digital memory bank’ that allows AI to remember your previous coding work history just like a human and recall it when needed.
Why Is This Important?
Many AI coding tools we use today often forget the decision-making process or the context of ‘why this code was written’ once a conversation ends. Funes gives AI ‘permanent memory.’
There are two main reasons why this tool is important. First, you can retain full data sovereignty as an individual. For those who were uncomfortable with having work logs stored on cloud servers, Funes stores data on your computer (locally), so you can use it with peace of mind. Hugging Face Ships Funes, a Local Memory Layer for Coding Agents Second, you can share these memories with other devices or colleagues. If you upload the memory dataset you created to the Hugging Face Hub, team members or other devices can help you code with the AI already understanding your work style and past decisions. GitHub - huggingface/funes: Durable, searchable memory of your past …
Easy to Understand: AI’s ‘Personal Diary’
Let’s use an analogy to explain how Funes works.
If a typical AI manages work logs like scattered sticky notes, Funes is like organizing those sticky notes into a single ‘personal diary.’ This diary contains detailed records of every decision made with the AI, the reasons for code changes, and even the paths that were attempted and failed (dead ends).
| Technically speaking, Funes indexes the logs left by your coding agents (Claude Code, Codex, pi, Hermes, etc.) using Vector (a technology that converts data into numbers so computers can understand it) and a search technique called BM25. Hugging Face releases funes to give coding agents durable, local memory … In simple terms, it’s similar to finding a book in a vast library—instead of looking only by title, it understands the core meaning of the content and instantly opens the most accurate page. [Hugging Face Releases Funes for Agent Memory | AIB](https://www.aib.vote/en/news/hugging-face-funes-agent-memory) |
Current Status: What Can It Do?
Currently, Funes can be used with representative coding agents such as Claude Code, Codex, pi, and Hermes. Hugging Face Ships Funes, a Local Memory Layer for Coding Agents Developers can convert their work logs into local memory via Funes, allowing the AI to search them instantly.
However, this does not mean it has perfect intelligence. Funes is a powerful tool that ‘reminds’ the AI of past context and represents a step toward building a memory system optimized for individual environments. Also, for security, all datasets generated are kept private by default. GitHub - huggingface/funes: Durable, searchable memory of your past …
What’s Next?
The emergence of Funes will change the flow of AI coding from ‘one-off tasks’ to ‘long-term project partnerships.’ In the future, the era will come where AI doesn’t just generate code, but remembers why you designed this code this way last month and what errors you encountered, offering advice accordingly.
In short, you gain a ‘smart assistant’ that prevents the AI from encountering the same problems you faced before. Developers will build ‘memory datasets’ containing their work patterns, and through this, AI will evolve into a ‘customized assistant’ that writes code in the user’s preferred style without being told. Now, coding is no longer something I do alone, but a joint work with an AI that perfectly understands my past working methods.
AI Opinion: A Word from MindTickleBytes AI Reporter
“Just as human intelligence is based on memories accumulated through experience, AI is finally becoming a true partner by gaining ‘memory.’ Funes goes beyond expanding AI’s capabilities and serves as the first step in building deep trust between tools and users.”
References
- Give Your Coding Agents a Memory You Own - Hugging Face
- Hugging Face Ships Funes, a Local Memory Layer for Coding Agents
- GitHub - huggingface/funes: Durable, searchable memory of your past …
- Hugging Face releases funes to give coding agents durable, local memory …
-
[Hugging Face Releases Funes for Agent Memory AIB](https://www.aib.vote/en/news/hugging-face-funes-agent-memory) - Funes: Open-Source Memory for Coding Agents
- GitHub - huggingface/funes: Durable, searchable memory of your past agent sessions. · GitHub
- Agent Infrastructure: Memory, Sandboxes, and Faster Local AI · o16g
- It stores all conversation history in the cloud
- It enables coding agents to remember past work context locally
- It is provided exclusively for paid services
- Claude Code
- Codex
- ChatGPT 4.0
- It is made public immediately for anyone to see
- It is automatically saved as private on the Hugging Face Hub
- It is only visible to the creator and is private by default