Agentbrain

AgentBrain

Self-evolving AI cognitive architecture with semantic memory, emotion modeling, personality traits, skill learning, and reward systems.

Features

Dual-Layer Memory

• Vector Memory: fast semantic similarity matching via local embeddings.
• Graph Memory: entity/relationship recall with semantic traversal.
• Long-term context preservation across sessions.
• Automatic memory consolidation and decay.

Adaptive RAG

• Query-aware retrieval instead of fixed top-k memory injection.
• Vector recall for semantic similarity, graph recall for entities and links, procedural recall for workflows, and reflective recall for lessons.
• Relevance criticism before prompt injection: filter stale, weak, unrelated, or contradicted memories.
• Memory reinforcement: useful recalls increase access count and confidence.
• Noise gating: greetings, teasing-only turns, and one-word reactions are not saved unless the response contains concrete durable findings.

Emotion System (AffectCore)

• Generative discrete emotions via cognitive appraisal theory.
• Valence/Arousal/Dominance tracking.
• Spontaneous mood drift based on interoception.
• Vietnamese sentiment detection.

Personality & Traits

• Real-time personality influence on responses.
• Warmth, assertiveness, directness, protectiveness scales.
• Self-aware persona management.
• Adaptive tone based on relationship depth.

Knowledge Extraction

• Automatic fact/entity extraction from sessions.
• Relationship type inference.
• Knowledge graph building.
• Cross-session semantic linking.

Tools

AgentBrain exposes these OpenClaw tools when the plugin is active:

agentbrain_status
agentbrain_emotions
agentbrain_graph
agentbrain_memories
agentbrain_skills
agentbrain_reflect

Storage

Brain state persists in ~/.openclaw/data/agentbrain/brain.db:

• SQLite database with vector embeddings.
• Relationship graphs.
• Memory decay tracking.
• Personality snapshots.