persistent-kb
Local, SQLite-backed cross-session memory for agents to store and retrieve lessons, decisions, and facts via FTS5.
skill install https://www.promptspace.in/skills/persistent-kbWhat it does
Persistent KB provides your AI agent with a long-term, local memory using a SQLite-backed knowledge base. It allows agents to save cross-session facts, lessons learned, and project decisions into a structured database that survives context window resets and machine reboots.
Why use this skill
Unlike standard LLM "memory" which is often session-bound or cloud-hosted, this skill offers a privacy-first, developer-centric approach. It solves the "amnesia" problem where agents forget technical hurdles or design decisions made in previous chats. By using FTS5 (Full-Text Search), it enables fast retrieval of relevant context without cluttering the current prompt window.
Supported tools
- SQLite: Local storage with high-performance FTS5 indexing.
- Python 3.10+: Zero-dependency implementation (standard library only).
- CLI: A robust script (kb.py) for easy integration with any shell-enabled agent.
The Output
The skill produces structured summaries of stored knowledge, including entry IDs, categories (lessons, decisions, references), and tags. It handles conflict resolution by superseding old facts, ensuring the agent always retrieves the most current information while maintaining history.
Use cases
- Avoid repeating technical research by storing findings as 'lessons'.
- Track architectural decisions across long-running development projects.
- Save complex commands or references to bypass repetitive documentation lookups.
- Build a searchable database of project-specific context and edge cases.
- Reduce token usage by retrieving specific facts rather than pasting full logs.
Example
Prompt
Sample output preview is available after purchase.