Agent Settings
The Agent Settings page in the Web UI (/agent-settings) configures the built-in AI agent itself.
This is broader than workflow type: agent or workflow type: chat configuration.
It controls:
- The Web UI assistant
- The model registry used by the built-in agent
- Tool permissions and bash command policy
- Provider-native web search for agent sessions
- The default soul/personality
- Prerequisites for Workflow Operator on Slack and Telegram
If you are looking for workflow YAML configuration, see Agent Step and Chat & LLM.
Start Here
The settings surface is split into smaller pages:
- Models & Providers for enabling the built-in agent, adding models, choosing a default model, and understanding model fields
- OpenAI Subscription for connecting a ChatGPT subscription and using the
openai-codexprovider - Tool Permissions & Bash Policy for deciding which tools the agent may use and how bash commands are filtered
- Personality & Web Search for selecting a default soul and enabling provider-native web search
What This Page Does Not Replace
- It does not replace workflow
type: agentsettings in DAG YAML - It does not replace workflow
type: chatsettings in DAG YAML - It does not configure chat-bot platform tokens for Slack or Telegram
Those are separate features with their own documentation.
Recommended Setup Order
- Enable the built-in agent
- Add at least one model
- Set the default model
- Review tool permissions and bash command policy
- Optionally select a default soul
- Optionally enable provider-native web search
- Only then configure Workflow Operator (Chat Bot)
Relationship to Workflow Features
Workflow type: agent
The settings page does not replace the agent workflow step. The agent step is a separate DAG feature for running a multi-turn tool-calling loop inside a workflow. See Agent Step.
Workflow type: chat
The settings page also does not replace the chat workflow step. type: chat is a workflow feature for direct LLM calls in DAGs. See Chat & LLM.
Workflow Operator (Chat Bot)
Workflow Operator depends on the built-in AI agent being configured first. Slack and Telegram conversations are routed into agent sessions, so the agent must already have a working model and policy configuration.
See: