Creating an Agent
Creating an agent is a four-step flow: mode, basics, knowledge, publish. The first two are required to save the agent; the last two can be done iteratively.
Open the Studio
Go to Agents in the sidebar and click New agent.
Pick a mode
Choose Standard for chat-first agents and Realtime for low-latency voice. You can change this later, but the prompt and voice settings reset.
Fill in the basics
| Field | What it does |
|---|---|
| Name | The friendly name customers see — e.g. Ari, Jordan. |
| Primary language | The default language for greetings and replies. |
| Welcome message | A one-line greeting; appears at the start of every conversation. |
| What should the agent do? | A plain-English description. Omniflow generates a starter system prompt from this. |
| Model | The LLM. Defaults to GPT-4o-class; switch in AI Infrastructure for the workspace default. |
| Voice | (Realtime only) The TTS voice — Aria, Puck, Sage, etc. |
Test in the sandbox
Omniflow opens a test console next to the editor. Talk to the agent (or chat with it) and watch the transcript and tool calls populate. Iterate on the prompt until it feels right.
Connect knowledge
Open the Knowledge tab and attach articles, FAQs, or documents. The agent retrieves the most relevant chunks at runtime — no manual prompt-stuffing required.
Add tools
Open the Tools tab to expose actions the agent can call: look up an order, transfer to a human, schedule a meeting, hit your own webhook.
Publish
When you’re happy, click Publish. The agent becomes available to channels and routing rules.
Publishing a Realtime voice agent does not connect a phone number. That happens in Telephony integrations. Until you wire up a channel, the agent only runs in the test sandbox.
Iterating after publish
Every change creates a new version. To roll back, open Versions and click Restore on the version you want — the live runtime picks up the change within a few seconds.
Open in Omniflow
Related
| If you want to… | Go to |
|---|---|
| Tune prompts and tone | Prompts & Tone |
| Pick a voice | Voice Models |
| Connect knowledge | Knowledge Sources |
| Add custom tools | Tools & Actions |