- Category
- AI Agents
- Rank
- No. 510Tools index
- Listed in
- #3 Give an agent memory
- Pricing
- Open Source
- Type
- TOOL
- Builder
- agentscope-ai
- GitHub
- 3.4k stars
- Latest release
- v0.4.1.11
- Date
About
AgentScope's memory-management kit for AI agents — episodic, semantic, and procedural memory stores with retrieval and consolidation.
What it does
ReMe gives AI assistants a shared, local knowledge workspace built from ordinary files. It captures conversations and text resources, creates daily notes, distills reusable knowledge, and retrieves it through keyword search, optional vectors, and links between related notes.
Why it's ranked here
The strongest reason to consider ReMe is control without giving up automation. People can inspect, edit, move, back up, and rebuild their knowledge base, while agents handle capture, indexing, consolidation, provenance links, and recall through several integration surfaces.
What's good
Markdown remains the source of truth, while indexes, graphs, and caches are rebuildable derived state. Search combines BM25, optional embeddings, and progressive link expansion. Consolidation distinguishes new, supporting, refining, and corrective evidence, and failed integrations remain eligible for retry.
Tradeoffs
The complete consolidation flow needs an LLM, and embedding retrieval requires explicit configuration. ReMe handles durable knowledge after conversations occur, not current context compression, summary injection, or tool-output trimming. Streaming jobs are available over HTTP but are not registered as MCP tools.
How to use it well
Use ReMe for assistants, coding agents, or research workflows that need inspectable memory across sessions. Start with local files, BM25 search, and links, then add embeddings or automated consolidation when justified. Keep session context management and decisions about proactive interruptions in the host agent.
Technical notes+
README.md describes the local service, Studio interface, CLI workflow, optional embeddings, and Markdown workspace model. pyproject.toml defines the reme-ai Python package, Python 3.11 minimum, Apache-2.0 license, setuptools build, reme console entry point, and core, development, and full extras. docs/en/framework.md documents configuration-driven Application, Job, Step, and Component layers, plus HTTP and MCP exposure rules. docs/en/auto_dream.md specifies changed-file extraction, digest integration, topic generation, checkpointing, and retry behavior. docs/en/auto_link.md details digest-only candidate recall and provenance-preserving wikilinks. docs/en/proactive.md confirms that proactive reading exposes previously generated topics without calling an LLM or deciding whether to interrupt users.
Observed
- License
- Apache-2.0
- Primary language
- Python
- Python requirement
- Python 3.11 or newer
- Packaging
- Published as reme-ai with setuptools; installable with core, development, or full extras
- Interfaces
- CLI, HTTP, MCP, embedded Python, and a local web workspace
- Platform support
- Declared operating-system independent
- Storage model
- User-owned Markdown, YAML, JSONL, and resource files with rebuildable indexes, graphs, and caches
Read from README.md, pyproject.toml, docs/en/auto_link.md, docs/en/framework.md, docs/en/proactive.md, docs/en/reme-blog.md, docs/zh/auto_link.md, docs/zh/framework.md, docs/zh/proactive.md, docs/zh/reme-blog.md, docs/en/auto_dream.md.
What it can do
Store episodic memories for AI agents
Event data and temporal sequences → Organized episodic memory records
Store semantic memories for AI agents
Knowledge and factual information → Structured semantic memory database
Store procedural memories for AI agents
Task procedures and skills data → Procedural memory patterns
Retrieve relevant memories based on queries
Memory search queries and context → Filtered relevant memory items
Consolidate memory data across different stores
Multiple memory store contents → Unified and optimized memory structure
Manage memory storage capacity and optimization
Memory usage parameters and thresholds → Optimized memory allocation and cleanup
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.