Models & Providers
This page covers the parts of /agent-settings that decide whether the built-in agent can run and which model it uses.
Enable Agent
The Enable Agent toggle turns the built-in Web UI assistant on or off.
This is also exposed through the environment variable:
export DAGU_AGENT_ENABLED=trueWhen the agent is disabled, the built-in chat UI is off and the Models table is hidden in the current Web UI.
Models Table
When the built-in agent is enabled, the page shows a Models table and an Add Model action.
The table shows:
NameIDProviderModel- Credential status
- Which model is the default
From the row menu, you can:
- Edit a model
- Set a model as the default
- Delete a model
Default Model
One model can be marked as the default. The Web UI agent uses that model unless the user overrides it in the chat UI.
The workflow type: agent step also falls back to this default model unless the step explicitly sets agent.model.
Add / Edit Model Fields
The current Web UI form includes these fields:
| Field | Meaning |
|---|---|
Name | Human-readable display name |
ID | Stable internal identifier used by Dagu |
Provider | Backend provider implementation |
Model | Provider-specific model identifier |
API Key | Credential for hosted providers; optional for local unless your proxy requires one |
Base URL | Custom endpoint override |
Description | Optional UI description |
Context Window | Optional model metadata |
Max Output Tokens | Optional model metadata |
Input Cost / 1M tokens | Optional pricing metadata |
Output Cost / 1M tokens | Optional pricing metadata |
Supports Thinking | Optional capability metadata |
Reasoning Effort | Optional default thinking depth. Shown only when Supports Thinking is enabled. Stored as low, medium, high, or xhigh; empty uses the provider default. |
ID Behavior
On model creation, the UI auto-generates the ID from the Name unless you override it.
In the current edit form, the ID is read-only after the model is created.
Provider Choices in the Current UI
The current Web UI provider dropdown offers:
AnthropicOpenAIOpenAI CodexGoogle GeminiOpenRouterLocalZ.AI
If presets are available, the create form also shows Import from Preset.
Credential Handling
The model form currently supports two credential modes:
- Direct credential providers:
anthropic,openai,gemini,openrouter,local, andzaiuse theAPI Keyfield.localkeeps this optional. - Subscription-backed provider:
openai-codexhidesAPI KeyandBase URLand instead shows a connection card for the OpenAI Codex login flow.
For openai-codex, the current Web UI requires an active connection before the model can be saved. The table's Credential column shows Subscription when that provider is connected.
For the subscription-backed OpenAI Codex path, see OpenAI Subscription.
Local Models
For Ollama and other local servers, choose the Local provider in the UI.
Important details:
Base URLis a base prefix, not a full endpoint- If
Base URLis left empty, Dagu defaults tohttp://localhost:11434/v1 - Dagu's local provider appends
/chat/completions - Dagu does not call Ollama's native
/api/generateendpoint from this path
That means these values are correct for a typical local Ollama setup:
Provider:LocalModel: your installed model tag such asqwen3.5orqwen3.5:latestBase URL: empty, orhttp://localhost:11434/v1
See Local AI for exact endpoint behavior and troubleshooting.
What These Fields Mean in Practice
Not every saved field is guaranteed to become a provider request parameter in every runtime path.
For example:
Base URLchanges where requests are sent- Pricing fields are stored as model metadata used by Dagu
Reasoning Effortis stored as model metadata and used only when the request path enables thinking mode for that model- Some fields are capability or UI metadata rather than provider-native knobs
When the exact runtime behavior matters, use the provider-specific documentation for that path. For local models, see Local AI.