REDDIT · REDDIT// 22d agoOPENSOURCE RELEASE

Metacog adds nervous system to AI agents

Metacog is an open-source framework that replaces passive agent memory with real-time "proprioceptive" signals and reinforcement tracking. Designed for Claude Code, it monitors an agent's internal state—such as token velocity, error loops, and action diversity—to prevent the repetitive failure patterns common in passive retrieval systems.

// ANALYSIS

Metacog solves the "Passive Librarian Problem" by pushing operational awareness to the agent instead of waiting for a retrieval query it doesn't know it needs.

–Five distinct "senses" detect anomalies like token velocity spikes and consecutive similar errors.
–A unique reinforcement model rewards the absence of failure, preventing the pruning of effective rules.
–Nociceptive interrupts force agents to state operating assumptions when stuck in loops.
–Minimal token overhead ensures the system remains silent and cost-effective until a deviation occurs.
–Per-project scoping allows for learning codebase-specific patterns without global behavioral contamination.

// TAGS

metacogclaude-codemcpai-codingagentopen-source

DISCOVERED

22d ago

2026-03-21

PUBLISHED

22d ago

2026-03-21

RELEVANCE

9/ 10

AUTHOR

richardbaxter