We decoupled memory from the model. Living Brain is one persistent brain for ChatGPT, Claude, Cursor, Codex and your own agents. Write in one, retrieve in any. 92.2% on LoCoMo. 464ms median retrieval. Up to 22.6x fewer tokens per question. LLMs have parameters. Living Brain has plasticity.

Your agent keeps its brain when you change its harness. Connect OpenClaw, Hermes or your own agent through the API, SDK or MCP. Switch frameworks and the accumulated context comes with you. The brain lives outside the runtime.

Same for models. Save something in ChatGPT. Pull it up in Claude. Keep going in Cursor, Codex or Grok. The model becomes interchangeable compute. The memory stays yours.

Why "Living Brain"? Brains keep what matters, link it to what they already know, and recall it when it's relevant. That's neuroplasticity, and it's what we designed around. 4 scoring policies decide every memory: ① Worthiness: is it worth keeping? ② Placement: where does it belong? ③ Resonance: what does it connect to? ④ Re-entry: does it matter for this task? The goal: the most useful information per inference token.

Memory that lasts has a new problem: it goes stale. Every Living Brain reports 3 health scores: • Freshness: how current is it? • Strain: where is maintenance building up? • Readiness: how usable is it right now? You see outdated sources before your agent uses them.

The numbers, from our internal evals: • 92.2% answer accuracy on LoCoMo • Up to 22.6x fewer tokens per question vs full context • Up to 95% lower cost per task • 464ms median retrieval • 3.7s median end-to-end answer We measure correct answers per dollar, retrieval and generation included.

Your brain doesn't have to live in an IDE. Use your brain across Claude Code and Codex. This workflow below costs less than $1 to run. Start with $5. API: $8/1M input, $0.50/1M output tokens. Bring your own model. Keep your own brain today:

The moment we've be waiting for ! So proud of the team!

let's go!!

As someone who constantly switches between AI tools for research, I've experienced firsthand how frustrating it is to lose context and start from scratch. This is amazing!

thank you @HeyChelseaLi for being a part of our @livingbrain journey!

truly amazing, congrats on your launch! are there any limits to Living Brain's memory?

we are preparing for the day when someone truly breaks its limits!

Thanks for sharing this, @ladyxtel . One persistent brain across ChatGPT, Claude, Cursor, Codex and your own agents is exactly what’s needed. The numbers look strong.🔥

excited to see you build with it!

Brilliant 🔥

thanks @JulianCorbett for the amazing support!

One of the things that breaks workflows down for many people is the lack of persistent memory across agents and sessions. This is definitely what we need more of right now. Great work, Christel! You and your team are addressing a core issue with this innovation.

thanks very much. we're just getting started! lots more to build.

A living memory system enables a good harness to leverage agents' persistent capabilities. Cool work, and thanks for sharing your results.

thanks @omarsar0 , huge fan of your work!!

this is literally my biggest pain and friction every time i switch from codex to claude or try out a new app or try to bring context from these apps to others like grok bot genuinely excited for this

try it today with codex and claude etc! will be great to see it in grok bot too

Memory go ☝️

more mem, more ram

know a guy who built a macos workspace for running fleets of coding agents where the long-lived ones keep their own memory and context. sent you a dm about an intro.

totally impressive.

thank you!

Incredible work as always!!! Always cheering you on.

love you!!

The brain provides context, it’s so important to keep the agent relevant

Nice, can i just plug this into my existing agents and get started?

yes, please try at

il futuro

This really useful and I’ve been waiting for this! Will check it out later

楼口某

Stellar project

Omg... this is huge. Thanks for this.

good solution for keeping context on each session

Can it sync memory across all agents instantly?

When Cursor starts a new chat the prior context is gone. Drop a .cursorrules file at the repo root with the decision log so every session replays the same ground truth. Reference it from your system prompt and the hallucination rate drops on refactors.

3.7s end to end with retrieval included should keep my agent replies feeling snappy also love this part, writing in ChatGPT and pulling it up in Claude will just saves so much time.

our @chatgpt plugin but i love how it works so well across codex and claude code as well

how's it handle conflicting info from different models? curious about that interaction.

context is king

precisely