Use local models in the same AI workspace
A local AI workspace should make local models useful, not isolate them in a separate app. ai.diy connects Ollama, LM Studio, and other OpenAI-compatible endpoints to the same chat, Canvas, knowledge, and model-selection workflow.
- Local providers
- Ollama, LM Studio, and custom endpoints
- Workspace
- One thread can move between configured models
- Privacy
- Local inference stays with the endpoint you choose
Ollama and LM Studio are provider choices
ai.diy treats a local model server as a provider. In Settings, select Ollama or a custom OpenAI-compatible endpoint, confirm the base URL, discover the models that are available, and choose one in the composer.
This keeps local experimentation next to cloud models instead of forcing a separate interface for each backend. Your last model choice is retained per provider, so switching away and back does not have to reset the workflow.
What stays local
If the selected provider is a local endpoint, the model request goes to that endpoint. The browser also keeps workspace state, local knowledge, Canvas artifacts, and settings in browser storage.
Local models do not automatically make every enabled tool local. Web search, remote MCP, cloud backup, and other connectors can send data to their own services when you enable them. Review the tool boundary before using sensitive material.
Container and network considerations
When ai.diy runs in Docker and Ollama runs on the host, the container may need a host gateway such as host.docker.internal. The exact address depends on your operating system and network layout.
Start with a simple text request, then verify vision, tool calling, and context limits for the chosen model. Those capabilities come from the model server and model, not from the label “local”.
A practical setup path
- 1
Run the local model server
Install and start Ollama or LM Studio separately, then confirm the endpoint responds on your machine.
- 2
Add the provider in ai.diy
Open Settings → API Keys, select the local provider, use model discovery, and test the connection.
- 3
Choose the model in the workspace
Use the provider and model controls in the composer. Keep web tools disabled when the task must remain entirely offline.
Questions people ask
Can ai.diy use Ollama models?
Yes. Ollama is a supported local provider. Connect the Ollama endpoint in Settings, discover the models it exposes, and select the model you want to use in the workspace.
Does ai.diy only work with local models?
No. Local endpoints and cloud providers can coexist. ai.diy is a BYOK workspace, so you can compare a local Ollama or LM Studio model with an authorized cloud provider without changing applications.
What is the main limitation of local AI?
Local model quality, speed, context length, and tool support depend on your hardware and the model server. ai.diy does not hide those differences; its role is to provide the workspace and provider switch.