ReMe is a local-first memory layer for AI agents that converts conversations and resources into Markdown memory files with indexing and retrieval features.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
ReMe is a local-first memory layer for AI agents. It turns conversations and resources into file-based long-term memory, then indexes, links, and consolidates memory for future recall.
How it works
Memory is stored as Markdown files with frontmatter and wikilinks. It progressively processes raw conversations and resources from session/ and resource/ into daily/, then consolidates into digest/ for long-term memory. It combines BM25, wikilinks, and embeddings for hybrid retrieval, and supports automatic memory, resource, and dream workflows that generate memory nodes.
Getting started
Install from PyPI:
pip install "reme-ai[core]"
Install from source:
git clone https://github.com/agentscope-ai/ReMe.git
cd ReMe
pip install -e ".[core]"
Environment variables:
cat > .env <<'EOF'
# Optional: used only after embedding components are explicitly enabled in the config.
# EMBEDDING_API_KEY=sk-xxx
# EMBEDDING_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
# Required for auto_memory, auto_resource, and auto_dream.
LLM_API_KEY=sk-xxx
LLM_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
EOF
Start the service:
reme start
Default address: 127.0.0.1:2333. If port is occupied, use:
reme start service.port=8181
Check status:
reme version
curl -s http://127.0.0.1:2333/version -H 'Content-Type: application/json' -d '{}'
Memory demo:
reme write \
path=digest/wiki/quick-start-demo \
name="Quick Start Demo" \
description="A first ReMe memory node" \
content="# Quick Start Demo\n\nReMe stores agent memory as readable Markdown.\n\nRelated: [[digest/wiki/memory-as-file.md]]"
reme search query="agent memory markdown" limit=5
reme read path=digest/wiki/quick-start-demo start_line=1 end_line=20
Generated file is Markdown with frontmatter.
Recent releases
Latest releases include:
- v0.4.1.3 (2026-07-20): fix(proactive): expose topics in primary answer; feat(index): add bounded memory-aware batch processing.
- v0.4.1.2 (2026-07-20): chore(benchmark): remove longmemeval final answer review file; refactor(config): remove max_file_bytes limit from ba.
- v0.4.1.1 (2026-07-15): refactor(embedding): defer provider construction until first remote call; feat(index): add file size limits and overs.
- v0.4.1.0 (2026-07-14): feat(chunker): better json chunker; feat(lme): add cli execution and agentic search tooling.
- v0.4.0.8 (2026-07-08): feat(benchmark): add lme benchmark steps; refactor(embedding): update embedding model initialization and session storage.
Traction
GitHub statistics: 3258 stars, 282 forks, 32 open_issues.
Memory system
Memory is organized under workspace_dir with directories: metadata/, session/, resource/, daily/, digest/; memory nodes live under daily/ and digest/ as files with frontmatter and wikilinks.
License
Apache-2.0
Getting started (notes)
Python 3.11+ required. Embedding-based semantic retrieval is optional and requires enabling components in the config and embedding_store setup as documented in the memory search guide.