agentbench

Self-improving long-horizon LLM agent — ChromaDB strategy memory + failure analysis, Grok-4 teacher labels → QLoRA-distilled LLaMA-3.2-1B student. 90% on Tau Bench, 95% inference cost reduction.

PECS: 基于 LangGraph 的四角色多智能体任务求解框架（Planner/Executor/Critic/Synthesizer）。WebShop 真实环境 +25pp (25% vs 0%)；GAIA 官方 53 题 26.4% vs ReAct 24.5%（McNemar 不显著）。含 AST 沙箱、50000 token 硬预算、FastAPI 限流/混沌/CI/Prometheus。