Skip to main content
stagewise ships with a curated list of built-in models, but you can define additional models that route to any configured endpoint — built-in or custom.

​
Adding a custom model

1

Open Models & Providers

Navigate to Settings → Agent → Models & Providers and scroll to the Custom Models section.
2

Click Add model

Click Add model to open the model editor dialog.
3

Configure the model

Fill in:
  • Model ID — The identifier your endpoint expects (e.g. llama-3.3-70b)
  • Display name — A friendly name shown in the model selector
  • Description — Optional description of the model’s strengths
  • Context window size — Maximum token context (default: 128,000)
  • Endpoint — Which endpoint to route this model to
  • Extended thinking — Whether the model supports chain-of-thought reasoning
  • Capabilities — Toggle supported input/output modalities

​
Capabilities

Each custom model can declare what it supports: Input modalities:
  • Text (always on)
  • Image
  • Video
  • Audio
  • File (PDF)
Output modalities:
  • Text (always on)
  • Image
  • Audio
  • Video
  • File
Tool calling — Whether the model can invoke tools. Required for file editing, workspace search, browser control, and other agent capabilities.
If you disable tool calling, the agent cannot perform code edits or interact with the browser. It becomes a chat-only model.

​
Custom model with a custom endpoint

A typical workflow for self-hosted models:
  1. Create a custom endpoint pointing to your local server (e.g. http://localhost:11434/v1 for Ollama)
  2. Create a custom model with the correct model ID for that server
  3. Select the custom model from the model picker in the chat sidebar

​
Provider options and headers

Advanced users can specify raw provider options (JSON) and custom HTTP headers for a model. These are passed directly to the AI SDK and can control provider-specific behavior like reasoning effort, thinking budgets, or streaming options.