Ollama

Ollama runs open models on your own machine. Connect it to Hull as an OpenAI-compatible provider to keep your code local and work offline.

Website: ollama.com

What to expectCopy Anchor Link

Local models are much smaller than frontier cloud models. They're more likely to loop, misuse tools or produce syntax errors, and they're often slower. Short conversations and fewer MCP tools help.

For reasonable speed you need a GPU with 24 GB or more of VRAM, or a Mac with 32 GB or more of unified memory.

Choose a modelCopy Anchor Link

Pick a model that supports tool calling and has a large context window, such as qwen3-coder:30b or devstral:24b. Browse the Ollama library for others. Smaller models can be enough for light tasks such as commit messages.

Set up OllamaCopy Anchor Link

1. Install Ollama from ollama.com and start it:

   Copy Text

   ollama serve

2. In another terminal, download a model:

   Copy Text

   ollama pull qwen3-coder:30b

3. Increase Ollama's context length. Ollama's default context is short and truncates prompts; agent work needs at least 32K tokens. Set it with the OLLAMA_CONTEXT_LENGTH environment variable before ollama serve, or with num_ctx in a Modelfile. See Ollama's FAQ.

Connect OllamaCopy Anchor Link

Set limit.context to the context length you configured in Ollama so Hull can manage the conversation. See Custom models for every model field.

TipsCopy Anchor Link

• Slow responses: local requests time out after 5 minutes by default. Raise provider.ollama.options.timeout (in milliseconds) if a large model needs longer.
• Remote Ollama: if Ollama runs on another machine, change the base URL to that host.
• See the Ollama documentation for installation and configuration.