Bỏ qua tới nội dung
TakoAPI
Quay lại kỹ năng

cortex-ai

bởi waynevaughan

Search & Research
—Chưa có đánh giá
Đánh giá của bạn

Tóm tắt

Persistent memory for AI agents by Sigma Labs.

Mô tả

Kỹ năng này làm gì

Cortex gives AI agents persistent memory across sessions using a file-based, git-tracked store. It combines a behavioral memory layer (Mind) with a document store (Vault) and hybrid search across BM25, semantic, and temporal signals. No LLM calls are required for storage or retrieval.

Every write is a git commit, making the full memory history auditable, diffable, and portable without an external database or LLM dependency.

Khi nào nên dùng

  • Retaining agent decisions and lessons between sessions
  • Searching past agent interactions by topic or recency
  • Auto-extracting entities and facts from conversation history
  • Versioning agent memory as git commits for auditability
  • Tracking user preferences across multiple agent runs

Quy trình ví dụ

ĐẦU VÀO

User asks: 'Remember that we prefer TypeScript over JavaScript for all new projects'

AGENT
  1. 1

    Observer extracts the preference from the conversation

  2. 2

    Mind stores it as a typed preference entry with reinforcement metadata

  3. 3

    Vault writes the entry to disk and creates a git commit

  4. 4

    Hybrid search index is updated for future retrieval

ĐẦU RA

Preference persisted as a versioned, searchable memory entry available in all future sessions

Cài đặt

clawhub install waynevaughan/cortex-ai

Chạy lệnh CLI, hoặc dán prompt vào Claude Code, Cursor, Codex hay coding agent bạn đang dùng.

Liên kết