Model settings

After you add a model in Open WebUI, you must flip two settings. The managed chat.yambr.com does this by default; self-hosters running the bundled compose stack also get it done via init.sh. If you install Open WebUI yourself, or you add a new model after the initial setup, check these by hand.

Setting	Value	Why
Function Calling	`Native`	Required for Computer Use tools to be invoked by the model. Without it, the model has no way to call `bash_tool`, `create_file`, etc. — it silently doesn’t.
Stream Chat Response	`On`	Enables real-time output streaming, including tool-call deltas. Without it, long-running tool sequences feel like the UI hung.

Set globally (recommended)

In Admin → Settings → Models → Advanced Params:

function_calling: native
stream_response: true

These become DEFAULT_MODEL_PARAMS — every model picks them up without per-model clicks. The bundled init.sh writes these on first Open WebUI startup:

POST /api/v1/configs/models
{
  "default_model_params": {
    "function_calling": "native",
    "stream_response": true
  }
}

Set per model

If you want different behavior per model, override in the model’s edit screen: Workspace → Models → [model] → Advanced Params.

Symptom: tools don’t fire

If the model replies in plain text every time without calling tools, 99% of the time Function Calling is Default instead of Native. The Default path relies on Open WebUI’s own tool-call parsing, which doesn’t play well with MCP tools wired through the Computer Use tool module.

Symptom: output hangs until complete

Stream Chat Response is Off. Turn it on; tool call output will appear as it’s produced.

Docker details Live browser

Getting started

Demo course

Yambr Platform

Self-hosting

Features

Skills

Integrations

Reference

Model settings

Set globally (recommended)

Set per model

Symptom: tools don’t fire

Symptom: output hangs until complete

​Set globally (recommended)

​Set per model

​Symptom: tools don’t fire

​Symptom: output hangs until complete

Set globally (recommended)

Set per model

Symptom: tools don’t fire

Symptom: output hangs until complete