HYPE OR FACT? · CHECKED 4 NOVEMBER 2026
Does Living Brain really give every AI one shared memory, at 92.2% on LoCoMo with 22.6x fewer tokens?
OUR VERDICT
Partly factOur original verdict: “The product is real but the numbers are all internal. Living Brain, from ChatAndBuild, is live in early access, with published docs, an MCP server, an API and pay-as-you-go pricing, so one memory store shared across ChatGPT, Claude, Cursor and Codex is a real offering. But every benchmark figure comes from the company's own internal tests, and the launch post puts the 92.2% LoCoMo score next to "up to 22.6x fewer tokens", which comes from a different benchmark (LongMemEval-S); on LoCoMo the company's own saving is 2.8x. We haven't tested it hands-on.”
THE CLAIM
“"We decoupled memory from the model. Living Brain is one persistent brain for ChatGPT, Claude, Cursor, Codex and your own agents… 92.2% on LoCoMo. 464ms median retrieval. Up to 22.6x fewer tokens per question."”
What’s actually true
- Living Brain is a memory layer from ChatAndBuild that stores knowledge outside the model and retrieves relevant context for each task. It's reached through a REST API, SDKs, a CLI and a hosted MCP server, and the docs say the API is in early access.
- The docs show MCP setup for Claude Code, Codex, Cursor and VS Code, plus ChatGPT and Claude connectors, so 'write in one, retrieve in any' matches how it's built.
- Company's figures, internal: 92.2% judged accuracy on LoCoMo across 1,540 questions using about 9.4K context tokens per question, against about 26K for full context (2.8x fewer). It reports 464 ms median retrieval and 3.7 s median total answer time.
- Company's figures, internal: on LongMemEval-S, about 5.1K tokens per question against a 116K full-context baseline (22.6x fewer), 95% lower cost per task and '0% observed hallucinations' under its scoring.
- The site itself says 'Results are from internal evaluations', that rankings apply only to the systems tested, and that savings depend on the workload.
- Pricing is published: pay-as-you-go API from $5 prepaid at $8 per 1M input tokens and $0.50 per 1M output tokens, with 1 GB storage and 100k requests a month. A team workspace comes with ChatChat Pro+, and Business is custom-priced.
- FourWeekMBA's independent write-up notes the same mismatch between the two benchmarks and found no independent evaluation.
Sources
Cost
Pay as you go: $8 per 1M input tokens and $0.50 per 1M output tokens, topped up from $5 with no subscription. A team workspace comes with ChatChat Pro+, and Business is custom-priced.
USE IT IF
People and teams who switch between ChatGPT, Claude, Cursor and Codex and want one project memory they all read and write. Also developers who want hosted memory for their own agents via MCP or an API.
SKIP IT IF
Anyone who needs independently verified memory benchmarks before they buy, or who isn't comfortable sending project knowledge to a third-party memory service. The API is still in early access.
How we knowChecked on 4 November 2026. Verdicts: Fact = the claim holds up; Partly fact = real, but with caveats or overstated parts; Hype = the headline claim isn't supported. Products change quickly, so check the maker's site before you buy. Our methods & sources ↗