Capy Cortex

Capy Cortex is an autonomous learning system that lets your AI assistant remember mistakes, extract patterns from sessions, and retrieve relevant knowledge to improve over time. It combines SQLite storage, full-text search, and TF-IDF ranking into a unified memory engine.

What Is This?

• Hook-based learning that automatically captures errors, corrections, and user preferences without manual intervention
• Two-stage retrieval system using FTS5 for fast candidate selection, then TF-IDF re-ranking for precision
• Consolidation pipeline that merges related rules into higher-level principles and applies confidence decay to keep knowledge current

Why Use It?

• Avoid repeating mistakes: Anti-pattern recording flags known failure modes during future interactions
• Fast, accurate retrieval: FTS5 indexing keeps searches fast even with thousands of entries; TF-IDF ensures semantically relevant results rank first
• Measurable improvement: Assistant adapts behavior based on extracted user preferences and consolidated patterns
• Low maintenance: Passive learning pipeline ensures continuous improvement with zero manual overhead

How to Use It?

Record a mistake and corrective rule:

capy_cortex.learn(
    event_type="mistake",
    content="Used append() inside a loop to build large lists",
    correction="Use list comprehensions or pre-allocated arrays for performance",
    confidence=0.9,
    tags=["python", "performance", "anti-pattern"]
)

Retrieve relevant knowledge before responding:

results = capy_cortex.retrieve(
    query="python list performance optimization",
    top_k=5,
    rerank=True
)
for entry in results:
    print(entry.content, entry.confidence_score)

Trigger session reflection after a completed interaction:

capy_cortex.reflect(
    session_id="session_20240815",
    depth="deep",
    extract_preferences=True,
    consolidate_threshold=3
)

Run manual consolidation across the knowledge base:

capy_cortex.consolidate(
    min_occurrences=5,
    similarity_threshold=0.82,
    decay_factor=0.95
)

When to Use It?

• Deployments with repeated interactions from the same user or domain
• Coding assistants where anti-pattern avoidance and preference tracking improve quality over time
• Any system where you want to trigger retrieve at response start, learn after user corrections, reflect at session end, and consolidate periodically (daily or every fifty sessions)

Important Notes

• Requires SQLite version 3.9.0 or higher; verify before deployment
• Domain-specific terminology may require custom analyzer via vectorizer_config parameter
• Avoid calling consolidate during active sessions due to full-table scans and re-vectorization latency

Try It in Happycapy

1. Open Happycapy in your browser—no install or signup needed to try.
2. Describe what you want in one sentence. Example: "I want an AI assistant that learns from my coding mistakes and remembers to avoid using append() in loops."
3. Capy Cortex will automatically capture errors during your session, reflect on patterns, and retrieve relevant anti-patterns the next time you ask similar questions.