This plugin provides an intelligent long-term memory system for AstrBot, enabling agents to store, retrieve, and manage conversational history and facts dynamically. It targets AstrBot users and developers looking to enhance AI agent capabilities with persistent, context-aware memory, improving interaction coherence and knowledge retention.

How It Works

The plugin employs a sophisticated hybrid retrieval system, merging BM25 sparse retrieval with Faiss vector retrieval, unified by RRF fusion. It utilizes a dual-route approach, maintaining both document and graph memory stores, each supporting keyword and vector search. Intelligent summarization, powered by an LLM, condenses conversation history into structured memories, differentiating between canonical and persona-specific summaries. Memory atomization treats each fact as an independent unit with a lifecycle, while a time-aware graph dynamically updates edge confidence and applies temporal decay, creating a robust, evolving memory architecture.

Quick Start & Requirements

Installation involves placing the plugin folder into AstrBot/data/plugins; dependencies are automatically handled by AstrBot. Configuration is managed via the AstrBot plugin configuration page, requiring embedding_provider_id and llm_provider_id (defaults can be used). The AstrBot official plugin Pages dashboard requires AstrBot version >= 4.24.2. Specific compatibility notes exist for Gemini and DeepSeek V4 LLM providers regarding tool call semantics.

Highlighted Details

• Hybrid Retrieval: Combines BM25, Faiss, and RRF for comprehensive search.
• Dual-Route Four-Mode Retrieval: Supports document/graph and keyword/vector search paths.
• Intelligent Summarization: LLM-based summarization into canonical and persona memories.
• Auto-Forgetting & Memory Atomization: Manages memory lifecycle, TTL, and decay.
• Time-Aware Graph: Dynamically updates edge confidence and temporal decay.
• Data Safety: Features automatic backups, rollback on index rebuild failure, and transactional deletion.
• WebUI Management: Offers a trilingual (zh/en/ru) dashboard for memory management and debugging.

Maintenance & Community

Support is available via GitHub Issues and a QQ Group (Password: lxfight). No specific contributors, sponsorships, or roadmap links are detailed in the provided text.

Licensing & Compatibility

This project is licensed under AGPLv3. This strong copyleft license may impose restrictions on use within closed-source applications or for commercial purposes, requiring derived works to be shared under the same license.

Limitations & Caveats

Users upgrading from versions v1.4.0-v1.4.2 may encounter data migration issues, requiring manual recovery steps by locating and renaming a backup database file.