recall

A personal semantic memory system. Store, search, and manage memories locally using vector embeddings.

Everything runs on your machine: FastAPI server, Qdrant vector database (Docker), and Ollama for local embeddings. Zero cost, full privacy.

Install

From PyPI

pip install recall-cli

From Homebrew

brew tap anelcanto/recall-cli
brew install recall-cli

From source

git clone https://github.com/anelcanto/recall.git
cd recall
./install.sh

Or the quick version:

make install

Prerequisites

• Docker — runs Qdrant vector database
• Ollama — local embeddings (brew install ollama && ollama pull nomic-embed-text)
• uv — Python package manager (brew install uv)

Quick start

# One-time setup (creates ~/.recall/.env, starts Qdrant)
recall init

# Start the API server (in a terminal, keep it running)
recall serve

# In another terminal
recall add "The quick brown fox" --tag test
recall search "fox"
recall list
recall status

CLI

recall init                                                       # Set up config + start Qdrant
recall serve [--host 127.0.0.1] [--port 8100] [--no-qdrant]     # Start API server
recall add "text" --tag work --source cli [--dedupe-key "..."]
recall search "query" --top-k 10 [--no-text] [--output table|json]
recall ingest <file> [--format lines|jsonl] [--source name] [--auto-dedupe]
recall list [--limit 20] [--cursor ...] [--output table|json]
recall delete <id>
recall status

Environment variables

Variable	Default	Description
RECALL_API_URL	http://127.0.0.1:8100	API server URL
RECALL_API_TOKEN	(none)	Bearer token for auth

Architecture

recall CLI  -->  FastAPI server (:8100)  -->  Qdrant (Docker :6333)
                       |
                       v
                 Ollama (:11434)
                 nomic-embed-text

• FastAPI serves the HTTP API
• Qdrant stores vectors and payloads
• Ollama generates embeddings locally using nomic-embed-text
• CLI talks to the API over HTTP

User config lives in ~/.recall/.env. Qdrant data persists in a Docker volume.

API endpoints

Method	Path	Description
POST	/memory	Store a memory
POST	/search	Semantic search
POST	/ingest	Batch import
GET	/memories	List with pagination
DELETE	/memory/{id}	Delete a memory
GET	/health	Service health check

Claude Code Plugin

recall ships as a Claude Code plugin — Claude can store and search your memories directly during conversations, with no manual CLI commands needed.

Install via Claude Code plugin marketplace (recommended)

/plugin marketplace add anelcanto/recall
/plugin install recall@recall

That's it. Claude Code handles everything else.

Install manually (MCP only)

If you prefer to wire up just the MCP server without the plugin system:

pip install recall-cli

Then add it to .mcp.json at your project root:

{
  "mcpServers": {
    "recall": {
      "type": "stdio",
      "command": "recall-mcp"
    }
  }
}

Prerequisites

recall serve must be running before Claude can use the MCP tools:

recall serve

Available MCP tools

Tool	Description
store_memory(text, tags?, source?, dedupe_key?)	Store a new memory. Returns ID.
search_memories(query, top_k?)	Semantic search. Returns scored results.
list_memories(limit?)	List recent memories.
delete_memory(memory_id)	Delete a memory by ID.
check_health()	Check if recall API, Qdrant, and Ollama are up.

Usage examples

Once connected, just talk to Claude naturally:

"Remember that I prefer using uv for Python projects"
→ Claude calls store_memory(...)

"What do you know about my React setup?"
→ Claude calls search_memories("React setup")

"Show me all my memories"
→ Claude calls list_memories()

"Forget the last thing you stored"
→ Claude calls delete_memory(id)

Check for updates

From the terminal:

# Refresh the marketplace index first
claude plugin marketplace update recall

# Then update the plugin
claude plugin update recall@recall

Or from inside Claude Code, use the /recall:update command to check for a newer version on PyPI and upgrade.

Configuration

The MCP server reads config from ~/.recall/.env (created by recall init) and respects the same environment variables as the CLI:

Variable	Default	Description
RECALL_API_URL	http://127.0.0.1:8100	API server URL
RECALL_API_TOKEN	(none)	Bearer token for auth

Development