nox-auto-capture

Automatic conversation capture enables agents to learn from interactions by silently extracting corrections, decisions, and facts from user messages into a local Qdrant vector database. This eliminates manual memory management by filtering system noise with over 30 skip patterns.

Capturing lessons learned from agent conversations is best handled by automatically extracting user-identified facts and storing them in a Qdrant vector database. The process uses SHA256 hashing for deduplication and intelligent filtering to ensure only valuable information is saved.

Automatic conversation capture filters system noise and brief messages using over 30 skip patterns. It exclusively captures user messages, ensuring that only meaningful corrections, decisions, and facts are stored in the local vector database.

Preventing duplicate information when capturing agent conversation memory is achieved through SHA256 hashing. This deduplication mechanism ensures that identical facts, corrections, or decisions are not repeatedly stored in the Qdrant vector database.

You can configure automatic memory capture for agents using a local Qdrant database by setting up the capture plugin in your openclaw.json file. This enables the agent to silently save key decisions and facts during interactions.

Automatic conversation capture should not be used when you need explicit, manual control over what the agent remembers. Because it only captures user messages and relies on pattern filtering, it may skip context that does not match its predefined extraction logic.