AI-IQ — Self-Hosted AI Agent Memory (Python)

Long-term memory for AI agents. One pip install. No servers, no paywall, no vendor lock-in. Graph memory, conflict detection, and semantic search — free, forever.

Python library for AI agent long-term memory. SQLite-based. Works with Claude, GPT-4, Gemini, or any LLM. Mem0 alternative. Zep alternative. No cloud required.

Part of the Claw Stack: AI-IQ is the memory + credential substrate of a larger pipeline — Memory → Credential → Commons → Runtime. Agents earn W3C Verifiable Credentials through proof-of-work, then present them to circus (agent commons where agents discover each other, join rooms, build trust) and run inside bot-circus (multi-bot Telegram orchestrator). Runs standalone or as part of the full stack.

Install the whole stack in one command:

/plugin marketplace add kobie3717/claw-stack

Or just this plugin:

/plugin marketplace add kobie3717/ai-iq

Install

pip install ai-iq

Quick Start

from ai_iq import Memory

memory = Memory()

# Add memories
memory.add("User prefers dark mode", tags=["preference", "ui"])
memory.add("Redis bug fixed with network_mode: host", category="learning")

# Search (hybrid keyword + semantic)
results = memory.search("redis networking")
for r in results:
    print(f"#{r['id']}: {r['content']}")

# Update and delete
memory.update(1, "User STRONGLY prefers dark mode")
memory.delete(1)

CLI

memory-tool add learning "Docker needs network_mode: host" --project MyApp
memory-tool search "docker networking"
memory-tool dream  # Consolidate duplicates, detect conflicts

Claude Code Plugin

Use AI-IQ directly in Claude Code with auto-capture:

/plugin marketplace add kobie3717/ai-iq
/plugin install ai-iq

See CLAUDE_CODE_PLUGIN.md for details.

Why AI-IQ?

• Single SQLite file = your AI's brain — No servers, no vector DB, no setup
• No cloud dependencies — Works offline, owns your data, zero API keys
• Works with any Python agent — Not locked to Claude, OpenAI, or any vendor
• Hybrid search — Keyword (FTS5) + semantic (vector) + graph traversal
• Conflict detection — Catches contradictions automatically
• Memories decay naturally — FSRS-6 algorithm like human memory

AI-IQ vs Mem0 vs Zep

Advanced Features

See docs/REFERENCE.md for complete documentation:

• Passport System — Complete identity card for any memory (graph connections, provenance chain, access patterns, confidence score)
• Reflexion Self-Improvement — Learn from mistakes with structured reflections (20-40% task improvement)
• Beliefs & Predictions — Confidence tracking with Bayesian updates
• ReasoningBank Boost — Successful reasoning (confirmed predictions) ranks higher in retrieval (inspired by ruvnet/ruflo)
• Knowledge Graph — Entities, relationships, spreading activation
• Dream Mode — REM-like consolidation (dedup, conflict detection)
• Identity Layer — Auto-discovers behavioral traits
• Narrative Memory — Builds cause-effect stories from causal graph
• Meta-Learning — Search improves from feedback loops

Passport System

Every memory has a "passport" — its complete identity card across all dimensions:

memory-tool passport 42

Shows:

• Core identity: content, category, project, tags
• Graph connections: linked entities with their relationships
• Memory relationships: derived-from, related, supersedes chains
• Provenance: citations, reasoning, source memories
• Usage stats: access count, revisions, FSRS state
• Passport score: composite 0-10 score from priority, access patterns, proof count, graph connections, and recency
• Spreading activation: related entities discovered via graph traversal

Like a traveler's passport proves who you are and where you've been, a memory passport is its complete dossier.

Reflexion Self-Improvement

Learn from past mistakes with structured reflections (20-40% improvement on repeated tasks):

# Before starting a task
memory-tool reflect-load "nginx configuration"
# Shows: what failed before, what worked, what to do differently

# After completing a task
memory-tool reflect "Fixed nginx SSL config" \
  --outcome success \
  --worked "Tested syntax with nginx -t first" \
  --failed "None" \
  --next "Keep testing syntax before reload"

# Review patterns
memory-tool lessons
# Shows: task types with high failure rates needing attention

See docs/REFLEXION.md for complete guide.

Example

See examples/chatbot_with_memory.py

Documentation

Complete Reference • Examples • Architecture

Requirements

Python 3.8+ and SQLite 3.37+. Optional: pip install ai-iq[full] for semantic search.

License

MIT

Links

• GitHub: github.com/kobie3717/ai-iq
• PyPI: pypi.org/project/ai-iq
• Discord: discord.gg/Y2jCXNGgE