Local-first macOS AI assistant. No telemetry, no cloud, no third-party AI. Everything runs on your Mac via Ollama, faster-whisper, and SQLite.

Fork of Omi with all cloud dependencies removed.

Ollami captures your conversations, transcribes them in real-time using faster-whisper on Apple Silicon, generates summaries and action items, and gives you an AI chat that remembers everything — powered entirely by local models.

macOS Swift App
      │
      ▼
Local Python FastAPI  (localhost:8080)
      │
      ├── faster-whisper    — speech-to-text (Apple Silicon optimised)
      ├── Ollama            — LLM for chat, summaries, action items, agent
      ├── SQLite            — conversation + memory storage
      └── Chroma            — vector search (local embeddings via Ollama)

All data lives at ~/.ollami/ — nothing leaves your machine.

git clone https://github.com/SpencerSmithSite/Ollami.git
cd Ollami

Step 1 — Build the desktop app

cd desktop
OMI_APP_NAME=Ollami ./run.sh

This compiles the Swift app and installs it to /Applications/Ollami.app. Only needed once (or after Swift code changes).

Step 2 — Launch everything

cd ..
bash scripts/start.sh

start.sh handles the rest automatically:

1. Checks that Ollama and Python 3.11+ are installed
2. Starts Ollama if it isn't already running
3. Pulls llama3.2 and nomic-embed-text if missing
4. Creates a Python virtualenv in backend/.venv and installs dependencies (first run only, ~2 min)
5. Starts the FastAPI backend on http://localhost:8080
6. Opens /Applications/Ollami.app

Logs are written to /tmp/ollami-backend.log and /tmp/ollami-ollama.log.

If you prefer to run the backend yourself without start.sh:

cd backend

# Create a virtual environment
python3.11 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# (Optional) copy and edit the config
cp .env.example .env.local

# Start the server
uvicorn main:app --host 127.0.0.1 --port 8080

The backend requires Ollama to already be running (ollama serve).

Copy backend/.env.example to backend/.env.local and uncomment lines to override defaults.

The Settings → Ollami panel in the app lets you change the Ollama URL, active model, Whisper model size, backend URL, and webhook plugins without editing files.

The backend exposes these routes at http://localhost:8080:

• Analytics — Mixpanel, PostHog, Heap, Sentry, LangSmith, Datadog
• Cloud auth — Firebase OAuth → local UUID token at ~/.ollami/token
• Cloud AI — OpenAI, Anthropic, Deepgram, ElevenLabs → local equivalents
• Payments — Stripe, subscription plans, upgrade prompts
• Remote storage — GCS → ~/.ollami/data/
• Mobile app — iOS/Android Flutter app (macOS-only fork)
• Firmware — Omi wearable and Omi Glasses firmware
• Cloud infrastructure — Modal serverless, Firestore, Pinecone, Typesense, Twilio

All phases complete.

• Phase 1 — Strip telemetry and cloud from Swift app ✅
• Phase 2 — Local Python backend (SQLite, faster-whisper, Ollama, Chroma) ✅
• Phase 3 — Plugin webhook support ✅
• Phase 4 — Benchmarks, export/import, launcher, tool-calling agent, local TTS ✅

See TODO.md for the full task list.

MIT — same as upstream Omi.