Living Brain memory layer cuts tokens 22.6x on LongMemEval-S, API from $5
testingcatalog · x · 2026-11-05
ChatAndBuild launched Living Brain, a persistent, self-learning interoperable memory layer that connects to ChatGPT and Claude, or plugs into Codex, Cursor and other agents via API, SDK, or MCP—compatible with 60+ AI platforms.
Key claims from internal tests (self-reported):
- Up to 22.6x fewer tokens per question on LongMemEval-S vs. pasting the full history (5.1K vs. 116K), with 95% lower cost per task
- Claims #1 on LoCoMo in internal comparisons: 92.2% judged accuracy, 94.7% on temporal queries, 0% hallucinations
- 464ms median memory retrieval, 3.7s median answer time
API access starts at $5 with no subscription. ChatAndBuild is part of the NVIDIA Inception Program.
More from coding & agent
- Fizgig 9.0.4: train Fibo 2.0 on a 12GB card, SAM 3.1 text-to-mask — shootthesound · 2026-11-05
- Single source of state could help coding agents — Redux comeback? — cnakazawa · 2026-11-05
- Nous Research launches Hermes Agent Plugin Hackathon with $20,000 in prizes — Teknium · 2026-11-05
- Alibaba's E-GenRM: generative reward model for open-ended shopping agents hits 87.4% accuracy — _reachsumit · 2026-11-05
- Microsoft adds 60+ native SharePoint ops to Work IQ, lifting agent pass rate from 70.1% to 84.6% — adnan_hashmi · 2026-11-05
- AI makes infinite software possible — and infinite data silos inevitable — lfschiavo · 2026-11-05