Claude-Mem

An AI that takes notes on other AI's work in real-time

114 followers

Transform ephemeral AI conversations in REAL-TIME as a permanent, searchable archive. Visualize development timelines, track decisions across commits, and collaborate with your team.
Free
Launch Team / Built With
AssemblyAI
AssemblyAI
Build voice AI apps with a single API
Promoted

What do you think? …

Alex Newman
Maker
📌
You know that moment when you open a new Claude session and realize you have to explain everything again? "So we decided to use SQLite because—" Stop. Claude doesn't remember. You're strangers again. That frustration is why I built claude-mem. AI assistants have amnesia. For quick questions, it doesn't matter. But real software projects span weeks. You make decisions on Tuesday that matter on Friday. You debug for three hours and finally find the root cause. That insight should persist—but it doesn't. Every session starts from zero. The naive solution was obvious: save conversations, search later. Tried it. Useless. Thousands of tokens of noise. What matters isn't the forty-seven files Claude opened—it's the moment you realized why the bug happened. Memory isn't storage. It's compression. So claude-mem watches Claude work, captures what matters, compresses it, and injects it into future sessions. Invisibly. You don't see memory forming—it just happens. Context arrives before you notice it was missing. The whole thing was built using itself. This comment? Written by Claude searching its own memories of the last ten weeks of development. Would love to hear how others are thinking about AI memory and continuity. What would you want an AI assistant to remember?
Masum Parvej

@az_ne This resonates with project work, what’s the most surprising thing Claude remembered that changed your process?

Alex Newman

@masump the most surprising thing is that it just “continues”

The surprise was that single sentences of what Claude did, in a timeline list, made such a massive impact on quality and performance overall.

Nima Aksoy

@az_ne Congrats on the launch Alex, this is sharp work. Upvoted. I am co founder of bestofweb .site and after your launch if you want more visibility you are welcome to join and introduce Claude Mem to our founder community.

André J

I usually store research conversations as markdown. And give it to an AI to then implement in the code. Is this what this sort of does, but more automatically? Or how doe sit work?

Alex Newman

@conduit_design no this takes notes in the background while you work.

You build with Claude code, and in background in real-time Claude-Mem uses anthropic’s agent-sdk to essentially run another Claude code alongside, that watches live, generates observations.

Then a timeline of the work u did plus id’s for the full observations so Claude can look stuff up if it needs to. But it barely ever needs to.

André J

Sounds very cool. Altho I dont full understand it. That said it might be because I use Cline. Cline makes a kind of task list in markdown as it works to keep track. might be related.

Alex Newman

@conduit_design This isn't a task list... but Cline is one that needs to be on my radar! :)

André J

@az_ne for sure! I sort of get it. Its sort of a context mem sidekick that helps you when you transition between context windows. I would probably get it if i used claude. But I never have. My flow is run cline in 3 ide's Let the agents work on the same codebase at the same time on differnt problems. For hard problems let them work on different branches on the same problem. Costs 100$ a day full throttle, but productivity is wild. I run cursor, vscode, and antigravity, but all of them with cline as the driver.

Chilarai M

Wow! This is really amazing idea.
So does it stay inside an IDE as an observer or inside a Git repo?

Alex Newman

@chilarai the plugin runs a worker that queues tool use responses and prompts, that generates observations

Worker is on :37777

There’s an http api in there

And then search-server is an mcp file that is routed to the http API to stay DRY and offer ability to use MCP anywhere

Chilarai M

@az_ne cool!

Preet Mishra

Pretty cool! How do I get in touch with you?

JS

Does this work with the browser version of Claude code?

Marcelo Martins

It's frustrating, indeed, having to cross your finger and hope Claude will stay on topic as it changes sessions. I've learned to keep a close on the tokens counts and to save my project requirements on an .md file. I after miss the sessions, though! Thanks for Claude-Mem, @az_ne ! You would think that those guys developing the Claude's LLM would have thought about this before, but they might be too busy optimizing the LLM.

New User

When using Claude-mem for testing, it seems that the token consumption is quite fast, right?