How does OpenWork work with Ollama?
OpenWork has a built-in Ollama screen. Start Ollama, open Settings › Ollama in OpenWork, pick or pull a model, and click Add to workspace. No API key and no account, and your prompts and files stay on your computer.
Updated
- Ollama address: http://localhost:11434
- LM Studio address: http://localhost:1234/v1
- No API key or account needed
- $0 OpenWork cost for local models
Set up Ollama in OpenWork
OpenWork checks each model for image support and thinking levels, and Sync all models matches OpenWork to everything installed in Ollama.
Install and start Ollama
Download Ollama from ollama.com and start it. OpenWork talks to it at http://localhost:11434.
Open Settings › Ollama
Wait for the status to show Ollama running. If it doesn't, start Ollama and click refresh.
Pick or pull a model
Choose a model you already have, or click Pull a model and type a name from the Ollama library. You can also pull it in a terminal first.
ollama pull qwen2.5-coder:7bAdd to workspace
OpenWork saves it as Ollama (local). Start a chat and pick the model in the model picker.
LM Studio, llama.cpp, vLLM, and other local servers
Any local server with an OpenAI-compatible API works. Add it to opencode.jsonc in your workspace folder, then run Reload OpenCode config.
Example: LM Studio
Replace the model ID with the one your server reports.
{ "provider": { "lmstudio": { "npm": "@ai-sdk/openai-compatible", "name": "LM Studio (local)", "options": { "baseURL": "http://localhost:1234/v1" }, "models": { "qwen2.5-7b-instruct": { "name": "Qwen 2.5 7B Instruct" } } } } }
Pick a model that can use tools
OpenWork is an agent: it reads files, runs tools, and calls MCP servers. Small local models chat well but often struggle with tools. If a task stalls, try a larger model or one built for tool calling, and keep a cloud model connected for long tasks.
Local and self-hosted models for a team
- Share one self-hosted model: run Ollama or vLLM behind an endpoint your team can reach, and add it once in OpenWork Cloud as a custom provider. It shows up on everyone's desktop.
- Route it through the AI Gateway to choose who can use it and count usage per person.
- Want only approved models? Admins can turn on Only models you provide, which stops people adding their own providers, including local ones.
Next steps
Full local model guide
Add a custom LLM or gateway
Share a self-hosted model with your team
Get the app
Frequently asked questions
- Is OpenWork free with Ollama?
- Yes. OpenWork is free and open source, and a local model adds no OpenWork cost. You only need a computer that can run the model.
- Does anything leave my computer with Ollama?
- Model inference goes to Ollama on your computer. Web tools, connectors, updates, and other enabled services can still make network requests; local inference is not a guarantee that the whole app is offline.
- Which Ollama model should I use?
- Start with a model built for tool calling, such as qwen2.5-coder. Bigger models handle multi-step tasks better.
- Does OpenWork work offline?
- Yes, once the model is downloaded. Chat on your files works without internet; web tools and online connectors don't.
- Why can't I add Ollama on my work computer?
- Your organization may allow only the models it provides. Ask your OpenWork admin.