Publishing
==========

Publishing a Voice Agent deploys it to the production infrastructure

.. image:: ../images/voice-publish.png
   :alt: Voice Agent publish flow — validate, review, publish
   :align: center
   :width: 100%

| so it can handle real calls. The publish workflow uses a review gate — an agent must be validated and approved before it can go live.

Publish Workflow
----------------

The workflow has four stages:

1. **Edit** — the agent is in edit mode (locked to the current user). Configuration changes can be saved freely.
2. **Validate** — the agent is validated to confirm all required fields are present and the configuration is consistent.
3. **Mark for Review** — after a passing validation, submit the agent for approval. The agent moves to *Pending Approval* status and becomes read-only.
4. **Approve / Reject** — a user with approval permissions reviews the agent and either approves (creating a new version and publishing) or rejects (returning it to edit mode with an optional comment).

Accessing the Publish Controls
-------------------------------

Click the **Publish** button (rocket icon) in the header toolbar. A popover opens with the current status and available actions.

The Publish button is disabled when:

* The canvas is in read-only mode (not in edit mode)
* Another user holds the edit lock
* You are viewing a historical version

Step 1: Enable Edit Mode
------------------------

Click the **Edit Mode** toggle in the header to take the edit lock. The header updates to show your name as the current editor.

Step 2: Validate
----------------

In the Publish popover, click **Validate**. The platform checks:

* All required configuration fields are filled
* The Twilio phone number is present
* The STT and TTS providers have their required credentials
* The LLM configuration is complete

If validation passes, a green checkmark with a timestamp appears. If it fails, an error toast lists the specific issues to resolve.

Step 3: Mark for Review
-----------------------

After a successful validation, click **Mark for Review**. The agent status changes to *Pending Approval* and the form becomes read-only. A notification is sent to workspace users with approval permissions.

Step 4: Approve or Reject
--------------------------

A user with the **Approve Voice Agents** permission sees **Approve** and **Reject** buttons in the Publish popover.

* **Approve** — marks the agent as approved, creates a new version snapshot, and publishes to EKS. The version appears in the version history dropdown.
* **Reject** — returns the agent to *Available* (edit mode). An optional review comment can be added and is displayed to the agent owner.

Publishing to EKS
-----------------

After approval, click **Publish → EKS** to deploy the agent to the Kubernetes cluster. A toast confirms success or reports an error. Once published, the agent is live and will handle calls routed to its configured phone number.

Phone Number Setup
------------------

For an inbound agent, configure your Twilio phone number's **Voice webhook** to point at the platform's inbound call endpoint. The endpoint URL is available in the **API Access** tab.

For an outbound agent, the platform dials numbers from a contact list uploaded via the **Schedule** tab. No webhook configuration is required for outbound.

Versioning
----------

Every approval creates a new numbered version (e.g. **v1.0**, **v2.0**). The version dropdown in the header lists all approved versions. Selecting a past version loads it in read-only mode. Click **Restore** to copy it back as the working draft.