Living Brain: Interoperable, efficient memory layer for agentic AI
ChatAndBuild has released Living Brain, a memory layer that keeps context outside any single AI model. Product Hunt's daily leaderboard for November 5, 2026 ranked the launch third that day, and the listing tags it for productivity, developer tools and artificial intelligence. Founder Christel Buchanan describes one persistent store that ChatGPT, Claude, Cursor, Codex and custom agents can all write to and read from. Each captured item becomes a page, and Living Brain links related pages as it compiles them.
Access runs through several paths. ChatGPT and Claude users can connect with their account sign-in and no API key. Claude Code, Codex, Cursor, VS Code, Hermes and OpenClaw connect to a hosted MCP server using a setup snippet, and developers can also build on the API or SDK. Existing ChatAndBuild plan holders already have access, and a trending X summary says pay-as-you-go pricing starts at $5.
The performance figures so far come from the company. It reports 92.2% judged accuracy on the LoCoMo benchmark, against 72.3% for a full-context baseline run on the model alone. Median retrieval takes 464 milliseconds and the median end-to-end answer takes 3.7 seconds. On LongMemEval-S, it cites about 5.1K tokens per question against 116K for full context, which is the basis for its claim of up to 22.6x fewer tokens. These are internal evaluations and have not been independently verified.
Why it matters
For teams building AI products, the central argument is that memory can outlive the model. If history sits in a separate layer, a user who moves from one assistant or agent framework to another no longer has to rebuild context, which lowers switching costs for end users. That same portability shifts lock-in toward the memory store, so builders should ask who owns stored memories and how they can be exported. Agent framework makers could also treat memory as a component to buy rather than build.
For investors, the token and cost figures are the economic hook. Sending only a few relevant memories instead of full histories reduces inference spend per task, and the company claims up to 95% lower cost per task. Those numbers rest on self-run benchmarks with the company's own baseline, so they are worth testing before they shape a valuation or a vendor decision.
Key points
- 1Living Brain launched on November 4, 2026, per FourWeekMBA's report of the company's announcement.
- 2Company-reported accuracy is 92.2% on LoCoMo, versus 72.3% for a full-context, model-only baseline.
- 3Median retrieval is 464 milliseconds, with a median end-to-end answer time of 3.7 seconds.
- 4Claims include up to 22.6x fewer tokens per question and up to 95% lower cost per task.
- 5Connects via a hosted MCP server, API or SDK, with ChatGPT and Claude available through account sign-in.
