friday

What it does

• Four MCP tools (add_memory / add_fact / memory_search / get_context) over stdio or HTTP, shared by Cursor, Claude Code, VS Code and friends
• Neo4j-backed directed graph: schema and service dependencies are traversable, so blast-radius questions beat string matching
• One-command Docker stack: installer verifies Docker, provisions containers, generates secure keys and prints MCP configs
• Neural Studio: a live 3D knowledge-graph UI for inspecting what your agents actually remember

When to use it

Teams whose sessions keep forgetting architecture decisions, schema rationale and environment quirks — and who want one self-hosted memory every IDE and CLI agent reads and writes, inspected through a graph UI rather than a pile of markdown.

Getting started

# option A: automated installer (verifies Docker, prints MCP configs)
curl -fsSL https://raw.githubusercontent.com/friday-memory/friday/main/install.sh | bash

# option B: manual docker compose
git clone https://github.com/friday-memory/friday.git && cd friday
cp .env.example .env
# fill in: FRIDAY_API_KEY, DEEPSEEK_API_KEY, MEM0_API_KEY, NEO4J_PASSWORD
docker compose up -d

Things to know

“Self-hosted” covers the containers, not the whole pipeline: the .env requires a DeepSeek API key and a Mem0 API key, so memory processing calls cloud services (the README cites DeepSeek at $0.14/M tokens). The graph is only as good as what agents write into it — four tools, no automatic capture. The 95% precision figure is the project's own benchmark badge, not an independent audit. MIT.