Recuris boosts long-horizon agent success via dual memory mechanism
omarsar0 · x · 2026-08-27
Recuris introduces a framework splitting agent memory into Working Memory and Experiential Memory to improve long-horizon tasks. Skill selection is grounded in current task state rather than full history, and a fixed Meta-Agent converts failed runs into validation-gated updates to Skill Memory. Across four benchmarks and ten models, it improved task success in 35 of 37 model-benchmark pairs.
Related event: Princeton's Recuris Boosts Long-Horizon Agent Success via Recursive Memory(2 posts)→
More from coding & agent
- Stanford's Self-Verification Boosts DeepSeek Past Claude — 机器之心 · 2026-08-27
- Claude iterates furiously on OpenSCAD code automatically — _Stocko_ · 2026-08-27
- Two vLLM recipes for Blackwell: NVFP4 KV cache buys 262K context and more streams — SeanHighness · 2026-08-27
- Agent Trading Demo: Links X API to iMessage for Inverse Cramer Strategy — MurrLincoln · 2026-08-27
- Paper Studies Agent Communication in Long-Horizon Competitive Settings — xuanalogue · 2026-08-27
- Zoetrope visualizes Claude Code sessions as live flow graphs — Saboo_Shubham_ · 2026-08-27