+15.33pp on SWE-bench Lite (n=150, 95% CI [+9.33, +22.00], p<0.05, paired bootstrap). Everything journalists, analysts, and partners need to cover REM Labs.
REM Labs builds persistent memory infrastructure for AI agents. The platform provides a simple 3-call API (remember, recall, forget), the Dream Engine — a neuroscience-inspired overnight consolidation system with 9 strategies and 5 depth levels — and delivers +15.33pp on SWE-bench Lite (n=150, 95% CI [+9.33, +22.00], p<0.05, paired bootstrap n=1000). Self-hostable. MIT license. No lock-in.
| System | Accuracy | |
|---|---|---|
| MemPalace | 96.6% | |
| AgentMemory | 96.2% | |
| Hindsight | 94.6% | |
| REM Labs | 94.6% | |
| Mem0 | 66.9% | |
| Zep | 64% | |
| GPT-4 | 53% |
REM scores 94.6% alongside Hindsight; MemPalace and AgentMemory lead the leaderboard. Full methodology and raw results at /benchmarks
Building REM Labs since 2025. The project started from the frustration of rebuilding RAG pipelines for every new AI application and watching them plateau at ~60% accuracy. The Dream Engine — inspired by how human REM sleep consolidates memories — was the architectural breakthrough that now delivers +15.33pp on SWE-bench Lite (n=150, 95% CI [+9.33, +22.00], p<0.05, paired bootstrap).
High-resolution screenshots available on request via press@remlabs.ai
The REM Labs logo is an orbital atom mark representing interconnected memory. Please do not modify, recolor, or distort the logo.
For press inquiries, interview requests, media assets, and high-resolution screenshots — reach out directly.
press@remlabs.ai