Docs / Local models
Local models
No account, no key, and no call leaving the machine, provided the model in Settings is a local one.
Ollama
Installing Ollama and pulling a
model that supports tool calling is enough on its own. Settings
defaults to http://127.0.0.1:11434/v1, and the
Detect button lists whatever is already pulled on the machine.
Why tool calling matters
A model without tool-calling support can still hold a conversation, but it cannot touch the board: writing blocks, connecting them, or reading a document all go through the same fifteen tools described in The canvas, and a model that cannot call them is limited to talk.
Any other local server
Ollama is the default, not a requirement. Anything that answers the OpenAI-compatible chat completions API works the same way, through the custom endpoint field in Settings. See Providers & endpoints for how that adapter is shared across every provider, hosted or local.