chaos-mind
par hargabyte
Notes & PKMRésumé
Hybrid search memory system for AI agents.
Description
Ce que fait ce skill
CHAOS Memory is a hybrid search memory system for AI agents that combines BM25 keyword matching, vector semantic similarity, graph relationship bonuses, and access-pattern heat signals into a single retrieval layer. Memories are stored and searched manually via a CLI. Auto-capture from session transcripts exists but is disabled by default.
Combining four retrieval signals in one query surfaces more contextually relevant memories than keyword or semantic search alone.
Quand l'utiliser
- Searching past architectural decisions before starting a new task
- Storing a pricing or product decision mid-session for future retrieval
- Looking up environment configuration details across agent sessions
- Anchoring a technical decision to specific code file locations via Cortex
- Scanning recent memories before a code review or handoff
Exemple de workflow
User asks: find any past decisions about authentication, then record today's choice
- 1
Run
chaos-cli search "authentication" --mode index --limit 10for a fast token-efficient scan - 2
Review returned memory summaries for relevant prior decisions
- 3
Run
chaos-cli search "authentication" --mode full --limit 3on the top hits for full detail - 4
Store the new decision:
chaos-cli store "Switched to JWT for stateless sessions" --category decision --priority 0.9 - 5
Confirm storage with
chaos-cli list 5
Agent surfaces prior auth decisions and persists the new one with high priority for future retrieval.
Prérequis
- Ollama installed locally (only required if enabling the opt-in auto-capture feature, available at ollama.com)
Installation
clawhub install hargabyte/chaos-mindExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.