memorai 🧠

Local-first AI memory system with semantic search. Built in Rust.

Store anything you want to remember, search it semantically, and generate user profiles — all running locally on your machine. No cloud, no subscriptions, no data leaving your box.

Features

• Semantic Search — Find memories by meaning, not just keywords
• REST API — Full HTTP API for integration with any app
• CLI — Command-line interface for quick access
• Profile Builder — Auto-generate user profiles from stored memories
• Bulk Import — Import many memories at once
• Embedded Database — SurrealDB runs inside the binary, no separate server
• Local Embeddings — Uses Ollama for embeddings, everything stays on your machine

Prerequisites

• Rust (latest stable)
• Ollama with an embedding model

# Install Ollama and pull the embedding model
curl -fsSL https://ollama.com/install.sh | sh
ollama pull mxbai-embed-large

# Optional: pull a chat model for profile generation
ollama pull qwen2.5:14b

Install

git clone https://github.com/iamngoni/memorai.git
cd memorai
cargo build --release

# The binary is at ./target/release/memorai
# Optionally, copy it to your PATH:
sudo cp target/release/memorai /usr/local/bin/

Usage

Start the API server

memorai serve

Server starts at http://localhost:8484 by default.

CLI Commands

# Add a memory
memorai add "Rust is my favorite programming language" --tags "tech,preferences" --source "conversation"

# Search memories
memorai search "what programming languages do I like?" --limit 5

# View stats
memorai stats

# Generate a profile
memorai profile

Configuration

All configuration is via environment variables:

Variable	Default	Description
MEMORAI_PORT	8484	API server port
MEMORAI_OLLAMA_URL	http://localhost:11434	Ollama API URL
MEMORAI_EMBED_MODEL	mxbai-embed-large	Ollama embedding model
MEMORAI_CHAT_MODEL	qwen2.5:14b	Ollama chat model (for profiles)
MEMORAI_DATA_DIR	~/.memorai/data	Database storage path

API Reference

Health Check

curl http://localhost:8484/health

Create a Memory

curl -X POST http://localhost:8484/v1/memories \
  -H "Content-Type: application/json" \
  -d '{
    "text": "I love building CLI tools in Rust",
    "tags": ["rust", "cli", "preferences"],
    "source": "conversation"
  }'

Search Memories

curl "http://localhost:8484/v1/search?q=programming+languages&limit=5"

List Memories

# All memories (paginated)
curl "http://localhost:8484/v1/memories?page=1&per_page=20"

# Filter by tag
curl "http://localhost:8484/v1/memories?tag=rust"

# Filter by source
curl "http://localhost:8484/v1/memories?source=conversation"

Delete a Memory

curl -X DELETE http://localhost:8484/v1/memories/{id}

Bulk Import

curl -X POST http://localhost:8484/v1/memories/bulk \
  -H "Content-Type: application/json" \
  -d '{
    "memories": [
      {"text": "First memory", "tags": ["test"]},
      {"text": "Second memory", "tags": ["test"], "source": "import"}
    ]
  }'

Get Stats

curl http://localhost:8484/v1/stats

Generate Profile

curl http://localhost:8484/v1/profile

Architecture

memorai
├── Axum HTTP server (REST API)
├── SurrealDB embedded (storage + indexing)
├── Ollama client (embeddings + chat)
└── Cosine similarity (vector search)

• Storage: SurrealDB in embedded/RocksDB mode — the database lives inside your binary. No external database server needed.
• Embeddings: Generated via Ollama's local API using mxbai-embed-large (1024-dim vectors).
• Search: Cosine similarity computed in Rust over stored embedding vectors.
• Profiles: Generated using Ollama's chat model (qwen2.5:14b by default).

License

MIT — see LICENSE