Ollama
Run Vibe Browser with 100% private, local models via Ollama. Your browsing data, DOM snapshots, and prompts never leave your machine.
Quick Start
1Install Ollama
Download and install from ollama.com. Single binary, no Docker or Python needed.
2Pull a Model
Browser automation requires strong reasoning and tool-use. We recommend Qwen 3.5 for the best experience.
# Recommended for Vibe Browser
ollama pull qwen3.5
# Lighter alternative
ollama pull llama3.1:8b
# Smallest, great for testing
ollama pull smollm2:1.7b3Connect Vibe Browser
Open Vibe Browser Settings, select Ollama (Self-Hosted) as your provider. It auto-connects to localhost:11434 and shows all your installed models.
Why use Ollama with Vibe Browser?
Absolute Privacy
Your browsing history, DOM snapshots, and prompts never leave your local machine. No cloud telemetry, no data harvesting.
Offline Automation
Automate intranet sites, local dev servers, and offline documents without needing an active internet connection.
Zero API Costs
Run unlimited automation tasks and process millions of tokens without worrying about usage limits or API billing.
Native Integration
Vibe Browser connects directly to Ollama's native API with full support for tool-calling capabilities out of the box.
Model Discovery
Vibe Browser automatically discovers models from your local Ollama instance. When you open the model dropdown in settings, you see all installed models directly queried from the /api/tags endpoint.
If you select a model from the recommended list that isn't installed yet, Vibe Browser automatically triggers a background pull via /api/pull to download it.
Recommended Models
| Model | Description |
|---|---|
| qwen3.5 | Default pick -- strong reasoning and tool use |
| llama3.1:8b | Meta's general-purpose model |
| deepseek-r1:8b | Strong reasoning and math |
| gemma3:4b | Google's efficient model |
| smollm2:1.7b | Smallest, runs on anything |
Model size and memory footprint vary by quantization and runtime settings. Browse all models at ollama.com/library.
Troubleshooting
Connection Refused
If Vibe Browser cannot connect, ensure the Ollama service is running:
curl http://localhost:11434You should see: Ollama is running. If not, run ollama serve to start it.
CORS Issues
If running Ollama on a different machine, set the OLLAMA_ORIGINS environment variable:
OLLAMA_ORIGINS="*" ollama serveThen update the Base URL in Vibe Browser settings to match your machine's IP (e.g. http://192.168.1.100:11434).
Model Runs Slowly
Memory requirements vary by model and quantization (for example Q4 vs F16). Try a smaller model ( qwen3:4b or smollm2:1.7b), close other apps to free RAM, or use quantized (Q4) variants for lower memory usage.