AI BriefingFoxclaw agent framework treats memory as a runtime with model-agnostic cognitive layer
·2026.11.10 22:04
Key point
Foxclaw is an open-source agent framework that treats memory as a runtime, achieving 1.0 flag_recall in offline drift detection evaluations.
Details
- Foxclaw is an open-source agent framework (MIT license) that treats agent memory as a runtime (HMR) rather than static storage to enable continuous learning and error prevention.
- Offline evaluations using all-MiniLM-L6-v2 achieved 1.0 flag_recall and 0.0 hard_fpr for drift detection, with self-learning tool weights converging from 0.3333 to 0.5262 over 8 rounds.
- The system uses a model-agnostic architecture where the cognitive layer (drift detection, policy learning) is independent of the LLM, allowing model swaps without losing accumulated judgment.
- A specific engineering fix replaced the default English embedding model with paraphrase-multilingual-MiniLM-L12-v2, raising Chinese semantic drift block_recall from 0.0 to 1.0.
- The codebase comprises 28,539 lines of Python across 181 source files, supporting agents for finance, office, and learning with integrated channel adapters for WeChat, Feishu, Telegram, and Discord.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
GitHub Next releases Ace, a local collaborative coding environment for humans and AI agents