lily-memory
by kevinodell
Coding Agents & IDEsBrief
Persistent memory plugin for OpenClaw agents.
Description
What This Skill Does
Persistent memory plugin for OpenClaw agents that stores and retrieves facts across sessions using a hybrid SQLite FTS5 keyword search and Ollama vector semantic search. Automatically captures facts from conversations and injects relevant memories before each LLM turn. Includes stuck-detection to break repetitive loops and deduplication on startup.
Combines keyword and vector search in a single lightweight plugin without requiring a dedicated vector database, while falling back gracefully to keyword-only mode when Ollama is unavailable.
When to Use It
- Remembering user preferences across separate agent sessions
- Recalling past decisions when an agent restarts mid-project
- Detecting when a conversation is looping on the same topic
- Looking up all stored facts about a specific entity or config
- Searching memory semantically when exact keywords aren't known
Example Workflow
User asks: What database did we decide to use for the API?
- 1
Extract keywords from the question
- 2
Run FTS5 keyword search against SQLite memory store
- 3
Run vector similarity search via Ollama if available
- 4
Merge and deduplicate results, inject top matches as context
- 5
LLM answers using injected memory facts
Agent recalls the stored fact 'api: database = PostgreSQL' and answers correctly without the user repeating themselves
Requirements
- Ollama with nomic-embed-text model (optional, for semantic search)
Installation
clawhub install kevinodell/lily-memoryRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.