Create an agent
Create agents from the Agents page or ask Vox to prepare a new-agent approval card. Every agent belongs to the active workspace and counts against that workspace's current plan limit.
Choose the channel first
| Channel | Current direction | Editor and delivery |
|---|---|---|
| Voice | Inbound or outbound | Voice editor; phone assignment and telephony runtime |
| Chat | Inbound | Chat editor; Website Chat deployment |
Channel is a structural choice. Chat agents remove speech configuration and cannot publish while a voice-only Call Transfer function remains attached. Create a separate agent when you need a different channel rather than treating channel conversion as a routine edit.
Creation paths
- Inbound voice template: creates a voice agent with the template prompt, voice, delivery preset, variables, and recommended-capability metadata.
- Outbound setup: creates an outbound voice agent from the entered identity and prompt fields.
- Chat template: creates an inbound Modular Website Chat agent with the selected prompt, model, variables, and template metadata.
- Vox: creates a paused voice or chat draft after inspecting prompt-authoring context and, for voice agents, the exact current voice catalog. Approval is required.
Recommended capabilities in a template are suggestions. They are not automatically connected, created, attached, or authorized. Configure and test each one explicitly.
Initial state
Normal completed creation produces a Paused agent. An incomplete older setup flow can leave an agent with status Draft and a Complete setup action.
Opening the editor initializes an editable version if one does not exist. Newly created agents do not become reachable merely because their configuration exists.
Limits and defaults
The server checks the workspace's live plan allowance before creation. All existing agent records in that workspace, including paused and draft agents, count toward the limit. Duplication uses the same limit.
When fields are omitted, the server uses channel-appropriate defaults. Current voice defaults include English, gpt-4o-mini, and Asteria for standard Modular voice creation; Chat removes voice and language fields. Templates and Vox can supply different valid values.
Use Compare plans and limits for the current allowance rather than relying on a hard-coded number.