MemoryAtlas

File-based long-term memory toolkit for AI agents — evolved from the MemoryScope project. Turns conversations and documents into readable, editable, searchable Markdown files linked by wikilinks. Three automated background processes (Auto Memory, Auto Resource, Auto Dream) progressively distil interactions into durable memory nodes and build wikilink relationship graphs over time. Runs as a local service (default port 2333) exposing a REST/MCP API; integrates with Claude Code via a first-party HTTP MCP plugin (v0.4.0.5) and with AgentScope-based agents via an in-process SDK adapter.

Storage
Markdown files with YAML frontmatter and wikilinks serve as the primary storage unit — every memory node is a human-readable, directly editable file. Auto Memory consolidates daily conversations into structured Markdown; Auto Resource ingests uploaded documents; Auto Dream further abstracts and links nodes into a connected knowledge base. Files live on local disk or any mounted path, organised under session/, daily/, digest/, and resource/ subdirectories.
Retrieval
Progressive hybrid search across three signals: wikilink graph traversal for relationship expansion, BM25 keyword matching, and dense embedding similarity. All three are fused (progressive widening via RRF) to return relevant memory nodes for a query. Exposed via a REST/MCP service API, a SKILL.md-based CLI integration, and a Python API (reme-ai package).
Self-host
Self-host: moderate
License
Apache-2.0
Pricing
Open-source Apache-2.0 Python package (pip install reme-ai[core]), free to self-host. Requires an OpenAI-compatible LLM endpoint and an embedding API (e.g., Aliyun DashScope). No hosted cloud tier found. · Free / OSS
GitHub stars
3,290
Last release
2026-08-10
Last commit
2026-08-10
First catalogued
2026-06-28

Strengths

  • Human-readable storage: every memory is a plain Markdown file — inspectable, editable, and version-controllable by both agents and users
  • Progressive hybrid search (wikilinks + BM25 + embeddings) without complex infrastructure — single pip install
  • Self-evolving knowledge base via Auto Memory / Auto Dream background processes; works across Claude Code (MCP plugin), AgentScope agents, and any SKILL.md-compatible harness
  • Apache-2.0, actively maintained with frequent releases (v0.4.1.3 as of 2026-07-20); evolved from MemoryScope with broader design and first-party Claude Code integration; now also ships a first-party Hermes Agent adapter

Watch out

  • Requires external LLM and embedding API keys (no fully-offline default path)
  • MemoryScope lineage branch still exists in the repo; ReMe and MemoryScope diverged — verify which is the authoritative successor for production use
  • No hosted cloud option: teams must provision and maintain the storage directory and API keys themselves
  • Service must be running before MCP/CLI integrations can be used; Claude Code plugin setup requires manual reme start
  • Marketing homepage (reme.agentscope.io) returned HTTP 404 on 2026-07-23 despite the domain resolving to agentscope-ai.github.io — treat the README on GitHub as the authoritative doc source until the homepage is confirmed live again

Best for

  • Personal assistants and knowledge-worker agents that need long-term memory stored in a human-readable, inspectable format
  • Workflows where both humans and agents read and write the same memory files (e.g., collaborative knowledge bases)
  • Claude Code users wanting persistent cross-session memory with zero infrastructure beyond a local Python service

How it integrates

Benchmark results

No sourced results yet.

Sources

Last verified 2026-08-01 · updated by refresh-framework-cards