Skip to main content
Agents is the product for building, testing and running voice agents. Omni is its realtime runtime. Use the visual builder, the API, or both to manage the same saved agent. The builder remains Live Beta; features, limits and production behavior may change during beta.

Build in the console

Describe the job, choose a voice, add knowledge and test your agent.

Build with the API

Create a saved profile with POST /v1/agents, then connect to /v1/omni.
Sandbox keys never bill. Live calls require available credit, and managed telephony is metered separately. See the pricing page for current runtime and managed Agents terms.

​
Console workflow

1

Open the Agent builder

Go to Launch an Agent. Sign up or sign in. The path is preserved through authentication.
2

Business and job

Start with the business name, then choose a starting role such as receptionist, support, scheduling or lead qualification. Edit the suggested job description to match your business. Use Custom role for your own workflow.
3

Knowledge and voice

Upload files, add website links, paste reference notes or attach existing organization knowledge bases directly in the creator. This step is optional. Existing KBs show source processing and failure counts. New files and links are staged until creation; remove them here to exclude them. Local files need to be reattached if you close the creator.Choose voice, speaking perspective and suggested capabilities. Capabilities still need configured tools before they can execute real business actions.
4

Review the editable brief

Edit the greeting and six instruction sections: identity and purpose, conversation flow, knowledge rules, actions and handoff, speaking style and boundaries. Voice delivery guidance is sent separately from business rules. Knowledge documents remain reference material and are not copied into the prompt.The starting templates are versioned. Use Preview refreshed template to inspect a replacement before applying it; Keep my edits preserves your current brief. Creation saves the reviewed instructions without a background rewrite. Click Create & test when ready.
5

Test before going live

Check any setup warnings first. Knowledge may still be processing after creation; wait for indexing before testing answers that rely on it. Run the browser greeting test. Confirm that the opening line, voice, and first response match your instructions. Confirm consent behavior when recording is enabled. Browser tests have no phone leg and cannot prove transfer, DTMF, hold, collect, or hangup behavior.
6

Put it on your website

Open Website & phone. The wizard drafts a Call Now snippet and publishes it when you added a public site URL (localhost is always included so you can preview). Add any remaining origins, publish if needed, and paste the script tag. Guide: Add your Agent to a website.
7

Connect a phone number

Assign an available managed number or connect a supported carrier route. Availability and regulatory onboarding vary by country. Agents usage and managed telephony are metered separately when PyAI carries the phone leg; see the pricing page for current terms.
Managed call-control execution is not available yet. An Agent can decide to emit transfer_to_human, send_dtmf, play_hold, collect, or end_call, but those carrier actions work today only when your own Twilio, FreeSWITCH, or SIP transport handles the corresponding Omni control frame. Do not enable a call-control tool on a managed number until the console marks managed execution available for that tool.

​
Move between the console and API

The Agents workspace has Your agents, Realtime playground and Build with API. A saved agent’s API integration tab provides read, update and realtime connection examples for that agent. You can share the editor URL, including its selected tab, with a teammate who has access to the same organization and project. Use a key from your console project when creating an agent through the API. The profile then appears under Your agents in that project. Edits through either interface update the same profile; new sessions load the latest saved configuration. Inline configure fields can override saved settings for one call. For direct application connections, use wss://api.pyai.com/v1/omni?format=pcm16&rate=24000&session_label=<agent_id>. Saved profiles remain optional: omit session_label and send inline configuration when you want to manage each session entirely in code. Existing Omni clients keep their current endpoint, scope and wire protocol.

​
Need implementation help?

Talk to AtomsAI, our AI services business, for white-glove setup, bespoke integrations and ongoing agent improvement. Services are optional, scoped and quoted separately from PyAI usage.

​
Next steps

Website widget

Publish a Call Now button with one script tag.

Create agents via API

The same profile, from code.

Knowledge bases

Hosted files, URLs, and pasted text.

Omni tools

Hosted, server, or client tools.