Agent settings
Every flow has a root agent that holds the shared configuration for the entire graph. Click Settings in the top toolbar to open it.
What root agent settings control
The root agent is the parent of all sub-agents in the flow. Its settings define the defaults that sub-agents inherit unless they override them individually.
| Setting | What it controls |
|---|---|
| Name | The display name shown on the Flows list and in the agent graph header |
| Description | A short summary of what this flow does — visible on the Flows list card |
| System prompt | The core instructions for the root agent. Sub-agents have their own system prompts; this one governs any logic the root handles directly |
| Model | The LLM provider and model used by default across the flow. Sub-agents that don't set a model override inherit this |
| Permission mode | How much the flow can do without pausing to ask for approval — applies to all sub-agents unless overridden per tool |
Choosing a model
The model selector has two dropdowns: first the provider (OpenAI, Anthropic, Azure, Google, and others), then the specific model from that provider's available list. A route summary appears below the selector confirming which provider will handle requests.
If the provider you need isn't listed, navigate to Model Providers from the link below the selector to add it.
Sub-agent model overrides
Each sub-agent can override the root model with its own provider and model choice. A specialist sub-agent that needs faster, cheaper responses can use a smaller model while the orchestrator uses a more capable one.
Permission mode
The permission mode is the most important operational setting on the root agent. It determines when the flow pauses execution to wait for a human decision.
| Mode | Behavior |
|---|---|
| Always ask | Every tool call pauses for your approval before running. The safest option while you're building and testing |
| Ask for actions | Read-only tools run automatically. Destructive and external tools pause for approval |
| Autonomous | All tools run without pausing. Use this for scheduled or unattended runs where no one is watching |
| Tools disabled | No tools are attached. The agent reasons and responds in text only |
Start with Always ask while you're designing the flow. Once you're confident in what the agent does, switch to Ask for actions or Autonomous depending on your use case.
Saving changes
Click Save settings at the bottom of the panel. Settings take effect immediately for new runs — in-progress runs are not affected.
Deleting a flow
The Danger Zone section at the bottom of the settings panel has a delete option. Deleting the root agent permanently removes the flow and every sub-agent it contains. This cannot be undone.