Skip to content

Latest commit

 

History

History
138 lines (106 loc) · 5.92 KB

File metadata and controls

138 lines (106 loc) · 5.92 KB

Intelligence & Memory Documentation

AI capabilities, knowledge management, and cognitive systems.

Core Intelligence

LLM & Cognition

World Model & Knowledge

Business Facts & Citations

Memory & Learning

Episodic Memory

Self-Evolution ✨ NEW

  • Self-Evolution & Reflection Pool - Agents learn from mistakes, generate skills, optimize capabilities
    • Reflection Pool (critique-based mistakes storage)
    • Memento-Skills (generate new skills from failures)
    • AlphaEvolver (optimize existing skills)
    • Integration with episodic memory and graduation

Agent Learning

Enhanced Governance (2026) ✨

Integration with Intelligence:

  • Episode data informs governance graduation decisions
  • Policy engine uses memory retrieval for context-aware decisions
  • Knowledge graph data validates strategic governance choices
  • See VALIDATION_METRICS.md for performance metrics

Canvas Intelligence

Canvas Summaries

Canvas State

Error Handling & Debugging

AI Debug System

Error Handling

Supervision & Multi-Level Learning

Supervision

Key Concepts

Memory Architecture

User Request → World Model → Episodic Memory → Knowledge Graph → LLM Response

Knowledge Types

  • Episodic Memory: Past experiences and outcomes
  • Business Facts: Verified truths with citations
  • Knowledge Graph: Entities and relationships
  • Formulas: Business logic and calculations

Intelligence Flow

  1. Input: User request or data event
  2. Retrieval: Memory + Knowledge + Facts
  3. Reasoning: LLM with retrieved context
  4. Learning: Store experience for future
  5. Graduation: Update agent maturity

Quick Links

By Feature

By Use Case

Performance

System Latency Notes
Episodic Recall ~10-100ms Temporal vs Semantic
Knowledge Graph ~50-80ms Local search
Reflection Pool ~50-100ms Vector similarity search
JIT Fact Verification <500ms With citation check
Canvas Summary ~2-3s LLM generation
Memento-Skill Generation ~30-60s LLM code generation
AlphaEvolver Optimization ~2-5min Per generation

See Also


Last Updated: April 12, 2026