Your AI Coding Tools Can Finally Remember — Privately

Developers increasingly lean on AI assistants such as Claude Code or Codex to write and maintain software. But these tools have a frustrating flaw: close the conversation and they forget everything. Each new session starts from zero, as if the project is brand new. The reasoning behind earlier choices disappears. A new open-source tool, funes, tries to solve that by giving AI assistants a durable, private memory.

Funes works by indexing the detailed transcript an AI assistant already leaves behind as it works — every search, mistake, documentation read, and change of direction. It turns those logs into a searchable local memory. Next time you open a session, the AI can automatically recall past decisions, name the exact session and turn where the reasoning appeared, and even open the full context if needed. You stop repeating yourself, and the AI can build on work from "last Tuesday" instead of meeting the project as a stranger.

Privacy sits at the center of the design. By default, everything is processed on your own computer — no cloud service analyzes your sessions, and no account is required. If you want memory to follow you across machines, you can optionally sync it to a dataset you own on Hugging Face, kept private by default. Before anything is published, the tool scans for passwords and secrets and withholds them.

The catch: funes is a tool for programming assistants, not general chatbots like ChatGPT. Installing it means running a command in a terminal, which will feel uncomfortable if you are not technical. It also cannot fix incorrect code, only help the AI remember past work. But for the growing number of people who build software with AI, this is a meaningful step: memory you control, evidence you can trace, and less time wasted redoing what was already figured out.