Quick Start (Self-host)

Not installed yet? Start with Installation.

Step 1 — Configure a LLM provider

Open SynapseOrch with the link the installer printed at the end — Synapse is running → http://localhost:3000/?token=…, which synapse start prints too. With login off it carries the access link's token, and synapse access-link prints it again.

If you chose an AI while installing, it is already set up — go straight to Step 2. Otherwise the app opens on Connect a model, which finds the model tools on this computer and takes a key (what it shows); or go to Settings → Models yourself.

Using Ollama (no API key required)

1. Install Ollama if you haven't already
2. Pull a model: ollama pull mistral
3. In Settings → Models, the Ollama (Local) card lists the models it found. Set Default model to ollama.mistral — an install that skipped the AI question already uses it — and click Save changes

Until Ollama answers with the default model, Chat's welcome line says Model not ready and why, with a link back to Settings → Models.

Using Claude, GPT or another hosted model

1. Click the provider's card — Anthropic Claude, OpenAI, Google Gemini, xAI Grok or DeepSeek — paste your API key and click the card's Save
2. Set Default model — e.g. claude-opus-5 or gpt-4o
3. Click Save changes

See Cloud Providers and Local Providers for every option.

The screens, step by step: Install on your computer, in the Guides.

---

Step 2 — Create an agent

1. Open Agents and press New agent
2. Fill in:
   • Name — e.g. Research Agent
   • Agent Type — Conversational
   • Tools — enable Web Scraper, or Browser Automation, to start
   • Instructions — e.g. You are a helpful research assistant. When given a topic, search the web and provide a concise summary with key facts.
3. Click Save

---

Step 3 — Chat with your agent

1. Open Chat and pick your agent from its name at the foot of the message box
2. Type a message: What is the Model Context Protocol and how does it work?
3. Watch the agent reason through the problem, call tools if needed, and return a response

---

What's next?