Local model · Local files · Visible tools
A local AI agent for Windows
LocalAI Cowork turns a Windows desktop into an agent workspace: define an outcome, select the permitted context and model, watch tool activity, and review the result.
What runs locally
The LocalAI Cowork application, its SQLite workspace data, task state, configuration, and local tool execution run on your Windows device. Credentials are stored through Windows Credential Manager. With Ollama, model inference can also run locally. The app does not require a LocalAI Cowork account and the current release has no telemetry sender.
What a local agent can do
A model becomes an agent when it can work with bounded tools and keep a task moving beyond a single response. LocalAI Cowork can provide selected file context, connect MCP servers, execute approved terminal operations, use reusable skills, and organize tasks into projects and pipelines. Approval policies can pause sensitive actions before execution.
Set up the local route
- Install Ollama and download a model that fits your available memory and task.
- Install LocalAI Cowork for Windows and select Ollama in the model settings.
- Start with a test project and only the files the task needs.
- Describe the expected output, inspect the proposed work, and review every changed file.
- Add MCP servers or remote providers only when the local setup cannot complete the task.
Local-first is a boundary, not a slogan
A remote model, web tool, remote MCP server, or connector necessarily receives data for the action you request. LocalAI Cowork makes those integrations configurable rather than mandatory. Check the policy of every external service, keep secrets out of prompts and project files, and use a local model for material that should remain on the device.
Hardware and quality
Local inference speed and output quality depend on the selected model, quantization, CPU, GPU, and available memory. A smaller model may be responsive but less capable at complex planning. A hosted model may produce stronger results but changes the privacy and cost boundary. LocalAI Cowork lets you make that trade-off per configuration.