memory-lancedb-pro plugin
A long-term memory plugin for the OpenClaw agent on LanceDB: captures preferences and decisions itself, hybrid search, noise fading
- 4,451 ★ on the repository
- Last commit
Broad access
Why this level
- Auto-capture reads conversation content and saves it to disk, and entries can include passwords and personal data
- Entry text is sent to the embedding provider, reranker and extraction model under your key
- The `delete`, `delete-bulk`, `upgrade` and `migrate` commands remove and rewrite the store
Install
- Works in
Manual install
openclaw plugins install memory-lancedb-pro@betaThe recommended route. After installing, bind the memory slot in openclaw.json and restart the gateway.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Auto-capture reads conversation content and saves it to disk, and entries can include passwords and personal data
- Entry text is sent to the embedding provider, reranker and extraction model under your key
- The `delete`, `delete-bulk`, `upgrade` and `migrate` commands remove and rewrite the store
This is third-party code. Review the repository files before installing.
What it does
memory-lancedb-pro plugs into OpenClaw as a memory plugin and keeps in LanceDB what the agent learns about you and the project: preferences, decisions, entities, events and worked cases. Auto-capture at the end of a turn extracts such facts from the conversation by itself (up to three per turn), and auto-recall puts up to three relevant entries into the prompt before the next reply. Search is hybrid: vector plus BM25 full-text, then cross-encoder reranking. Old and unimportant entries fade by a Weibull model while frequently recalled ones stay, and memories sit in three tiers: peripheral, working and core. Memory is split into scopes: global, per agent, per project or per user. The store is managed with `openclaw memory-pro`: list, search, export, import, bulk delete and migration.
Who it is for. For OpenClaw users tired of repeating the same preferences to the agent and re-explaining past decisions in every new session.
Good fit when
- The agent forgets your rules between sessions: tabs instead of spaces, error handling, the chosen stack
- You need an answer to "why did we pick PostgreSQL a month ago": decisions are stored in memory and found by search
- Several agents or projects must not share memory: the `agent:<id>`, `project:<id>` and `user:<id>` scopes isolate entries
- You want to index `MEMORY.md`, files in `memory/` and session records for semantic search with citations and line numbers
- You need your own store without a cloud: embeddings can come from Ollama, and a reranker is optional
Not a fit when
- You do not use OpenClaw: the plugin targets its plugin architecture and hooks from version 2026.3 onward
- Your CPU lacks AVX instructions: LanceDB's native vector search can crash with SIGILL unless you enable the mode without native cosine
- Your conversations hold data that must not go to outside services: entry text goes to the embedding provider and the extraction model
- You need a plain note in `MEMORY.md` with no vector database: files are enough for a small project
Example request
Remember: in this project we write tests with Vitest and run migrations only through PrismaLimitations
Requires OpenClaw 2026.3 or newer and a CPU with AVX support. Search needs an embedding provider: Jina, OpenAI, Voyage, Ollama or anything compatible with the OpenAI API. OpenAI is unavailable from Russia without a VPN, so take Jina, SiliconFlow or local Ollama. Auto-capture works only when the hook has `allowConversationAccess` permission. Release 1.1.0 is a beta and installs with the `@beta` tag. An MIT license badge is present but the repository has no LICENSE file. Before bulk operations the author advises copying the LanceDB directory instead of relying on export.
How to disable. Unbind `plugins.slots.memory` and set `enabled: false` in the `memory-lancedb-pro` entry of `openclaw.json`, then run `openclaw gateway restart`. The LanceDB store stays on disk; delete it by hand if you do not need it. If you installed with the setup script, it has an `--uninstall` mode.
FAQ
How do I install memory-lancedb-pro in OpenClaw?
Run `openclaw plugins install memory-lancedb-pro@beta`, bind the memory slot `plugins.slots.memory` to the plugin in `openclaw.json` and restart the gateway with `openclaw gateway restart`. An npm install exists too, but then you must give the plugin path as an absolute one in `plugins.load.paths`.
What does the plugin remember and how does it work?
After a turn, messages are sorted into six categories: profile, preferences, entities, events, cases and patterns. Duplicates are filtered in two stages, first by vector similarity, then by a model decision to create, merge or skip the entry.
Does it need a paid key?
It needs an embedding provider. The author recommends Jina; OpenAI, Voyage and Ollama also work. The reranker and the extraction model are optional; without them search falls back to cosine similarity and extraction to regular expressions.
What commands does the plugin have?
`openclaw memory-pro` offers `list`, `search`, `stats`, `delete`, `delete-bulk`, `export`, `import`, `reembed`, `upgrade` and `migrate`, plus OAuth sign-in: `auth login`, `auth status`, `auth logout`.
What do the /lesson and /remember commands do?
`/lesson` saves a lesson from the dialogue as two entries: a fact with the pitfall and cause, and a decision with a behavior rule. `/remember` stores the given text with a suitable category and importance and returns the entry ID.
Discussion
No questions or reviews yet.
Related
PluginsA long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Long-term memory for AI agents: a skill and the mindmemos command store facts from conversations and recall them in later sessions
Shared semantic memory for agents: conversations are saved as markdown and recalled by search across Claude Code, Codex and other agents
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
A long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
Long-term memory for AI agents: Markdown files as the source of truth, a SQLite index as a cache, shared by Claude Code, Codex and others
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| memory-lancedb-pro (this entry) | no | 4,451 | ||
| Mem0 | no | 68,007 | a memory layer with plugins for Claude Code, Codex, Cursor and OpenCode, while memory-lancedb-pro is built for the OpenClaw memory slot | |
| memsearch | no | 2,804 | shared semantic memory with markdown files on Milvus, while this one has LanceDB, entry decay and access scopes | |
| MCP Memory Service | no | not required | 2,054 | self-hosted memory with a knowledge graph and REST API via MCP, while the plugin lives inside OpenClaw and adds auto-capture |
README in short
The README starts with motivation: most agents have amnesia and the plugin turns OpenClaw into an assistant with memory. The quick start offers a community setup script and a manual install through `openclaw plugins install` or npm, with a configuration example. The memory architecture section explains that `MEMORY.md` files remain the source of truth while LanceDB serves as a semantic index over them. Then come the features: hybrid search, reranking, a multi-stage scoring pipeline, extraction in six categories, decay, scopes, auto-capture and noise filtering. Setup, embedding providers, commands, migrations, rules for AGENTS.md and hook adaptation for OpenClaw 2026.3 follow.