reme-ai

What it is and what it does

ReMe is a persistent memory system for AI agents that stores knowledge as ordinary Markdown files with frontmatter and wikilinks, keeping all data under user control in the local filesystem. It combines keyword search (BM25), optional semantic embeddings, and relationship traversal to retrieve relevant context without loading entire knowledge bases into memory. The system includes automated workflows—Auto Memory distills conversations into daily notes, Auto Resource imports external documents, Auto Dream consolidates notes into long-term knowledge, and Auto Link writes relationships back into files.

The package integrates with agent frameworks through CLI commands, HTTP endpoints, Model Context Protocol (MCP), and embedded Python APIs. It runs as a local service (default port 2333) and includes an optional web UI (ReMe Studio) for browsing and editing. Core file operations, BM25 search, and wikilink traversal work without LLM credentials; AI-powered memory evolution requires OpenAI-compatible API keys. The 18 runtime dependencies include fastapi for the service layer, pydantic for configuration, and loguru for logging.

Use it for:

• Give personal assistant agents (QwenPaw, OpenClaw, Hermes) a user-editable long-term memory layer that persists across sessions.
• Preserve coding style, project decisions, and workflow experience for coding agents like Claude Code across multiple development sessions.
• Build a searchable, traceable Markdown wiki from conversations and resources that both users and agents can maintain and query.
• Enable agents to learn from experience by recording successful procedures, failed attempts, and periodic reflections as indexed memory nodes.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

ReMe is a local-first knowledge base that converts conversations and resources into searchable, interconnected Markdown files with wikilinks, designed to give AI agents and users a persistent, editable memory layer.

Yes. ReMe is actively maintained, has no known vulnerabilities, uses a permissive license, and solves a concrete problem for agent developers: persistent, editable, searchable memory under user control. The low install friction and modular design (core features work without LLM keys) make it accessible. Install if you're building agents that need to retain and evolve knowledge across sessions, or if you want a local-first alternative to external memory services.

Install

reme-ai on PyPI

pip install reme-ai

uv add reme-ai

poetry add reme-ai

Installing reme-ai

Before you install

Low install friction with a pure-Python wheel; active maintenance (last commit 2026-08-13, 3311 stars). Requires Python 3.11+ and 18 runtime dependencies including fastapi, openai, and numpy—a substantial but standard stack for agent-memory systems.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions, making it suitable for both open-source and proprietary agent integrations.

Quickstart

pip install "reme-ai[core]"
reme start
reme write path=digest/wiki/demo name="Demo" content="# Demo\n\nReMe stores memory as Markdown."
reme search query="memory markdown" limit=5

Requires Python 3.11+. LLM-powered features (auto_memory, auto_resource, auto_dream) need LLM_API_KEY and LLM_BASE_URL environment variables; basic file operations and BM25 search work without them.

Verify before relying

• Performance characteristics and scalability limits for large knowledge bases (file count, search latency).
• Embedding model requirements and resource overhead when semantic retrieval is enabled.
• Compatibility and integration depth with specific agent frameworks beyond the mentioned examples.
• Data persistence and backup guarantees under concurrent access from multiple agents.

Package facts

Evidence: reme_ai-0.4.1.7-py3-none-any.whl

Keywords: llm, memory, agent, agentscope, ai, mcp, reme

Tags

More Artificial Intelligence packages

Further reading