cortex-ai skill

Overview

This skill provides persistent memory and structured recall for AI agents, built to capture, search, and manage knowledge across sessions. It combines a typed behavioral memory, a git-backed document vault, automatic extraction from conversations, and a hybrid search engine. The design is file-based and zero-LLM, focusing on auditability and reproducible memory state.

How this skill works

The system records memories in two complementary layers: a Mind module for behavioral, typed entries with decay and reinforcement, and a Vault for document storage where every write becomes a git commit. An Observer watches interactions and extracts salient facts automatically into the memory graph. Search uses a hybrid approach (BM25 + semantic vectors + temporal ranking) to return relevant results that respect entity tracking and time-based decay.

When to use it

• When you need persistent agent memory across sessions with audit logs.
• To maintain a versioned document store where every change is tracked via git.
• When you want automated extraction of facts from conversations into structured memory.
• If you require hybrid, time-aware search combining keyword and semantic relevance.
• For projects that demand zero-LLM storage and reproducible memory states.

Best practices

• Design typed memory schemas for predictable recall and easier consolidation.
• Tune decay and reinforcement parameters to reflect the correct memory lifespan for your domain.
• Commit meaningful changes to the Vault and rely on git history for auditing and rollback.
• Index documents with clear metadata (timestamps, entities, sources) to improve hybrid search relevance.
• Validate Observer outputs and provide feedback loops to refine automatic extractions.

Example use cases

• Customer support agent that recalls past resolutions, preferences, and follow-up commitments over months.
• Multi-session research assistant that aggregates documents, tracks key entities, and surfaces temporally relevant notes.
• Robotics or automation workflows needing durable decision logs and behavioral reinforcement history.
• Compliance-oriented apps that require auditable memory and a git-tracked document trail for reviews.
• Team knowledge base where conversations are auto-harvested into searchable, structured memory.

FAQ