beacon.sh

Open-source self-improving memory for AI coding agents

Open Source

Developer Tools

GitHub

SDK

Visit WebsiteSee on Product HuntGithub ⧉

Love the idea of carrying lessons across coding agents. How does Beacon handle it when two agents learn conflicting things about the same codebase?

The portability is the bit I keep thinking about. Most sessions end with one useful lesson that's tied to a specific harness — a Claude Code quirk, a Cursor setting — and another that's true for the whole codebase. When a memory like that gets carried into a different agent, does Beacon tag or filter it by where it came from, or does everything travel as-is? I'd worry a harness-specific trick could read as a general rule in the wrong context.

Congrats on the launch @justindsouza !

What happens when two branches need different fixes for the same issue? Does Beacon track which advice applies to each branch? Or could an agent end up using a fix that only works on main?

Congrats Justin. Saw the launch on X. How does Beacon decide which learnings are worth turning into reusable skills and how does it discard outdated or conflicting learnings?

Ours is one `CLAUDE.md` in the repo and `AGENTS.md` just points to it, so every agent reads the same file and a change goes through a commit like any other code. When I approve a memory in Beacon, can it land as a file in the repo so a teammate sees it in the PR?

This is like AI agents having Apple Continuity. Very cool.

Que: how do you handle the artifacts across agents?

Does it ever save the wrong context and bring it back later when you don’t actually need it?

Congrats @justindsouza & team!

Thanks for sharing @fmerian :)

Lovely product! absolutelt loved the thought behind it and the problem it is trying solve for people who build with ai. Best wishes @justindsouza and team :)

The portability angle is the part that hits home for me — I switch between Claude Code and Cursor in the same day, and right now the second tool starts from zero no matter what the first one figured out. Turning past sessions into reusable skills instead of another summary doc is the right shape for that. One thing I'm curious about: are those memories scoped per repo by default? A workaround that makes sense in one codebase could be actively wrong in another, so I'd want the compounding layer to know which project a lesson came from.

Nice! love the idea

i am always switching between Cursor and Claude so this is very helpful so they both have reusable skills and memories

Really cool. I jump between T3 Code, Cursor and Codex, so this is great. Also open source. 🔥

Losing context across chats is one of the most frustrating things I encounter.

And as such, Beacon looks like a promising tool in that direction.

All the best for the launch!

Love it. Makes tokens last longer without the friction of my agents needing to load the whole context from scratch.

Congrats on the launch @justindsouza! Finally I can switch between Claude Code and Codex without re-explaining the same context all over again 🙌

Love that memory travels across agents like Claude Code and Cursor, so every session makes the next one smarter instead of starting from zero.

Wow, this is something i needed from long time. Switch from AI tool to tool the memory of those tools are quite act differently. Beacon has been sorted this out. Contrates on your launch and ideas. Its really solves the problem that many have. I'll surely look deep into it.