memo-fast

Your AI agent forgets everything between sessions. memo-fast fixes that.

It indexes your codebase -- functions, classes, git history, dependency graphs -- into a vector database you own. Next time your agent opens the project, it picks up exactly where you left off.

"what changed while I was away?"     -->  memory_resume
"how does the auth flow work?"       -->  memory_query
"explain OrderProcessor in depth"    -->  memory_deepdive
"show me the architecture topology"  -->  memory_spatial_map

How it works

memo-fast runs as an MCP server. Your agent connects to it and gets 8 tools for searching, navigating, and understanding your codebase through semantic memory.

Your code is processed into embeddings and stored in a vector database you control. Pinecone, Qdrant, or ChromaDB -- your account, your keys, your data. memo-fast processes everything in-memory and stores nothing on its side.

MCP client (Claude Code, Cursor, etc.)
    |
    | 8 tools via MCP protocol
    v
memo-fast (extract + embed)  -->  Your vector DB

Tools

CLI = also available via the CLI method (see below)

Tool	What it does
memory_resume	Recovers context from recent git commits. Start every session here.
memory_resume_extended	Like resume, but also resolves the dependency graph of every touched entity.
memory_refresh	Context for files you are currently editing (max 10 files, 60s cooldown).	CLI
memory_query	Semantic search. Ask in natural language, get code, commits, and files.
memory_deepdive	Finds an entity by name, pulls its source and direct dependencies.
memory_spatial_map	Full dependency graph with PageRank ranking and community clustering.
memory_enrich	Generates semantic descriptions for code entities (improves search quality).
memory_index	Index files and commits (max 30 files per call).	CLI

Getting started

1. Subscribe at mcpize.com/memo-fast
2. Add your vector DB credentials (Pinecone, Qdrant, or ChromaDB) in the dashboard
3. Add memo-fast to your MCP client config
4. Open your project and ask your agent anything about the codebase

The CLI Method

For initial indexing and automatic updates, memo-fast includes a lightweight CLI. It reads your project files and git history directly from disk and sends them to the server for processing.

Install

# macOS (Apple Silicon)
curl -fsSL https://github.com/sys-2077/memo-fast/releases/latest/download/memo-fast-darwin-arm64.tar.gz | tar xz -C /usr/local/bin

# macOS (Intel)
curl -fsSL https://github.com/sys-2077/memo-fast/releases/latest/download/memo-fast-darwin-amd64.tar.gz | tar xz -C /usr/local/bin

# Linux
curl -fsSL https://github.com/sys-2077/memo-fast/releases/latest/download/memo-fast-linux-amd64.tar.gz | tar xz -C /usr/local/bin

# Windows
# Download memo-fast-windows-amd64.zip from GitHub Releases and add to PATH

Usage

memo-fast init                  # One-time setup per project
memo-fast index                 # Index all files + recent commits
memo-fast index --incremental   # Index only the last commit
memo-fast hook install          # Auto-index on every git commit

memo-fast init does not ask for API keys. Collection is derived automatically from the current repository name. The CLI reads your MCPize token from ~/.mcpize/config.json (create it with npx mcpize login). Vector DB credentials are configured in MCPize dashboard, not in the CLI.

Once the hook is installed, your memory updates itself after every commit. No manual re-indexing needed.

Credentials

You bring your own storage. Configure one of these backends:

Credential	Required	Description
VECTOR_BACKEND	No	pinecone (default), qdrant, or chroma
PINECONE_API_KEY	If Pinecone	Your API key
PINECONE_HOST	If Pinecone	Your index host URL
PINECONE_INDEX	No	Index name (default: memo-fast)
QDRANT_URL	If Qdrant	Your Qdrant Cloud URL
QDRANT_API_KEY	If Qdrant	Your API key
CHROMA_HOST	If ChromaDB	Your ChromaDB host

Data processing

memo-fast is stateless. Your source code is processed in-memory to generate embeddings, then written directly to your vector database. We do not store, cache, or log your code. See DATA-PROCESSING.md for details.

License to Use

See MCPize terms.