One place for everything your AI needs to know about your project
If you build anything big with an AI assistant, you've felt the gap. A real project isn't just code — it's folders of docs, months of decisions, meeting notes, the reasons behind the architecture, the timelines, the plans, and the things you tried that didn't work. To be genuinely useful on a project like that, an assistant has to understand the domain you're working in, not just autocomplete inside it.
That's the problem we've been working on with Neonmem.
Even when an agent has "memory," it's usually a flat pile of text it can search but can't put in order. It doesn't know which decision replaced an earlier one, what's a plan versus a settled fact, or where you stopped yesterday. As the project grows, so does the mess — more data, more decisions, more threads to keep straight.
We wanted one place that holds all of it, in order, so your agent works like a real code buddy: aware of the project, and aware of how you work. So we built Neonmem as a single-point memory — one local cartridge that carries your project's reasoning, structure, and direction, with an understanding layer that keeps it organized.
What's inside
-
A single memory cartridge. One local
.neonmemfile — not a vector database, but a connected graph of decisions, dead ends, rules, and the reasoning between them. Yours, on your machine. - An understanding layer. Neonmem doesn't just store facts, it organizes them — clustering the memory into subsystems and mapping how the project fits together, so your agent gets the shape of the whole thing, not just fragments.
- Resume where you left off. It keeps the open thread and the exact point you stopped, so a fresh session picks up instead of starting cold.
- The importer. Point it at your docs, your codebase, and your meeting notes, and it reads the lot — turning them into one connected memory: the decisions, the endpoints, the call graph, the plans. Run it again as things move and it updates the memory in place, merging what's new and skipping what it already knows. Your agent opens already knowing the project.
- Plans. Most memory only looks backward. Neonmem also tracks where you're going — the goals and next steps — so the agent knows the direction, not just the history.
- Local and private. Offline embeddings, no cloud calls, no third-party model reading your code. It runs as its own small, self-built brain.
Where we are
Neonmem is in public beta. It's genuinely useful today, but we're early, and a lot is on the way — team memory (sharing a project's cartridge across a whole team), more autonomous workflows where the agent acts more on its own, broader platform support (macOS and Linux), and steady improvements to everything above.
If you live in an AI assistant and you're tired of re-explaining your project every morning, give it a try and tell us where it falls short — that feedback is shaping what comes next.
Our paradigm is that AI here is to augment human not replace , human in control AI takes weights.
https://neonmem.com/ · Windows x64 · free for personal use
Top comments (0)