atmem

General-purpose memory, context governance & execution investigation for AI agents.
- pip install atmem

[atmem @AtMemAi]

AtMem 2.3.8 is out 🚀

Hermes one-command support, source-linked context, typed evidence, bounded retrieval, fail-closed sufficiency checks.
📈 LongMemEval-V2: AtMem: 5/23 vs AgentRunbook-R 4/23 & Mem0 1/23
📈 DolphinBench: AtMem: 9/30 vs Mem0 7/30
> pip install -U atmem==2.3.8
If you like it give us Star and Follow us on GitHub 🙏
https://t.co/9kY5jkGlMM

[atmem @AtMemAi]

We Fine-Tuned a Memory Classifier for AtMem. Here's Why We Didn't Ship It

https://t.co/3lvyPnoTeB

[atmem @AtMemAi]

Thanks for the feedback.
AtMem provides persistent memory for AI agents, with governance, audit trails, and activity tracking.
It integrates with OpenClaw, LangChain/LangGraph, Pydantic AI, and custom agents. It stores memories and retrieves relevant information for the agent’s context.
An agent can use AtMem’s MCP tools to explicitly remember or retrieve information. Supported adapters and plugins, such as the OpenClaw integration, automate memory capture and context injection once configured and enabled.

[atmem @AtMemAi]

https://t.co/02WqsKemRv

Please check AtMem presentation about agents and AtMem

[atmem @AtMemAi]

the atmem code will have two direction:
AtMem Structure: “What is connected to what in this repository?”
AtMem Experience: “What have we learned while working on it?”
So AtMem Structure + Experince give a better picture to coding agents

[atmem @AtMemAi]

Working on atmem code, adding memory to coding agents things are on track. Testing the effectiveness of it is hard...

[atmem @AtMemAi]

I think testing is hard but when it comes to coding we can reply commits ... it's my idea for testing. In these days we are doing more agentic

[atmem @AtMemAi]

@lostonwww on the otherhand, we are doing more agentic code development, like software factories etc. We also have a lot of open sources, now we can see how things can improve if we add memory or whatever tool in the outcome retrospectively ... it's just idea not done it yet

[atmem @AtMemAi]

@humansandaiboss great idea looking forward

[atmem @AtMemAi]

@lostonwww Fixed

[atmem @AtMemAi]

Exactly it’s hard to penetrate the crowded noisy space of coding agents in particular. I just see we need to bring coding memory to AtMem it is the use case at least for me and my team. We can play with it. There are good solutions out there but I’m after reinventing the wheel in this case

[atmem @AtMemAi]

@lostonwww Need to fix it

[atmem @AtMemAi]

@ajs6888 I agree it’s not full replacement for separate agents but can work if you create tools around cli

[atmem @AtMemAi]

Don’t overlook Coding agents like Codex/Claude/Github Copilot custom agents: built-in coding tools, multi-agent delegation, MCP support, and orchestration without separate LLM API keys.

[atmem @AtMemAi]

sometimes your workflow can sit in cli and use coding agent to achieve your goal, this way your agentic work is integrated in the code. It comes with its limitation but has its place.

[atmem @AtMemAi]

We won’t leave that solely to the agent. The plan is to carry repo/worktree, commit and dirty-state context, then have AtMem check applicability before recall. Branch names alone aren’t enough. Today’s preview conservatively withholds code-dependent memories after revision changes.

[atmem @AtMemAi]

today discussin with a friend, we tried to use atmem for coding memory, I found AtMem in Github Copilot can be use via MCP, it is not seamless but can be used. We discussed about whether we can have a flavour of AtMem coding agent agnostic and instead of MCP passive path uses hooks and can integrate actively in coding agents. So I'm targeting a special release for AtMem on Code, it shall support out of the box with adaptors for Codex, Claude Code, Muse Code, Grok and Github Copilot. I'll keep you posted on that.

[atmem @AtMemAi]

@humansandaiboss It’s a show off impressive one useful? Of course not

[GooGZ AI @PaulGugAI]

Do this at the lights in front of the cops, and see how long you last☠️

[Elon Musk @elonmusk]

Maybe we should start a competitor to Amazon 🤔

[atmem @AtMemAi]

Anthropic replaced Git with Scion. Why?

AI agents pushed code merges from 2,500 to 30,000+ per day, making traditional source control a bottleneck.

The result? ~100× faster checkouts and 200× less infrastructure overhead.

The next bottleneck in AI coding isn't generating code. It's the infrastructure handling it.

https://t.co/oZTCCQjBXG