Testing Your Companion
The test environment screen - every control, dialog, and state - and what to check before publishing.
The Test screen is the same conversation surface your visitors get on the public/share page, rendered for you as the owner. It's not a stripped-down preview build - it's the identical chat component (voice, text, camera, guides, tickets, and all) running with owner privileges instead of public-visitor restrictions. Anything that works, breaks, or looks off here is exactly what a real visitor would experience.
Test sessions are not private. A test conversation creates a normal chat record the same way a live visitor session does - it increments the companion's session count and shows up in Conversations and in the Dashboard tabs (Usage, Engagement, Analytics, Support) exactly like any other chat. The only thing private about test mode is that nobody else has the URL to that specific test chat - it is not excluded from your stats, and it is not hidden from your own dashboards.
Where It Lives
From the companion's settings page, click the Test button in the header. It opens /assistants/[assistantId]/test in a new browser tab. Each past test conversation gets its own URL at /assistants/[assistantId]/test/chats/[chatId], reachable from the chat history drawer described below.
Layout and Controls
The Icon Row
Above the input area, a row of pill buttons controls the session:
- History - toggles a chat-history drawer (only shown in test mode, or to authenticated users on a share link) listing your past test conversations, grouped by date (Today, Yesterday, This Week, This Month, then by month). Includes a search box that filters by conversation title or ID, infinite scroll, a refresh button, and an empty state ("Your conversations will appear here") when there are none yet.
- Resources - opens the companion info popup.
- Share - opens the full Share dialog for switching visibility and copying the public/share link. Not redescribed here - see that page for every field and toggle in it.
- Save - saves the companion to your own collection (same heart-toggle behavior as anywhere else in the app).
Starting a Session
Below the companion's carousel and name, the input bar lets you either type a message or start a voice session (a single pill shows "audio" or "text" depending on what the companion supports and whether you've started typing). If the companion has Quick Start Prompts configured, they appear as clickable suggestion chips once the greeting animation finishes revealing.
During a Session
Once connected, the icon row is replaced by the live chat input bar: mic toggle, assistant-audio mute toggle, cancel-response button, a camera toggle (only shown if the companion has the capture_visual_context tool/skill enabled - see Camera Mode below), and session timer/time-remaining indicators when the companion has a time limit configured.
The Companion Info Popup
Clicking Resources opens a dialog with four tabs: Guides, Media, Knowledge, and Links - the companion's published guides, its uploaded resources, its knowledge sections, and its link list, all read-only. If the companion has no published guides, the popup automatically switches to the Media tab on open (unless you've already picked a tab yourself). Picking a guide from the Guides tab launches it as a full-screen guide runner overlay, pausing the session's mic and inactivity timer while it's open.
Dialogs and Modals
Mic Permission Modal
Opens automatically the first time you start a voice session (or mid-session if the browser mic permission changes). It auto-requests browser mic access; if the browser denies it, the dialog switches to a Blocked view with browser-specific steps (Chrome, Safari, Firefox, or a generic fallback) to re-enable it, plus a Try Again button and a Dismiss option. Denial doesn't fail silently - testing without mic access is testing broken for every visitor without one either.
Connecting Modal
Shown while the realtime session is being established. It's a fixed loading screen (no close button - it dismisses itself once the session goes active) that cycles through contextual tips about the companion (its about text, resource-backed answers, image/video/guide capabilities) while a status message types itself out.
Data Collection (Intake) Modal
If the companion has data collection enabled with at least one custom field, this "Quick Check-in" form appears before the session starts. It renders exactly the fields configured on the Advanced tab (required vs optional, per-field validation), prefills anything you've already submitted before (from a local cache, keyed per companion) or that matches your signed-in name/email, and blocks Continue until every required field validates. Cancel closes without starting the session.
Ticket Review Modal
Appears when an Action block in a guide is about to create a support ticket. It shows the auto-drafted Subject (read-only), an editable Description, and an editable Email field (locked to your verified email - shown with a "verified" tag - if you're signed in, otherwise freely editable). Submit Ticket creates the ticket; Cancel discards the draft. Closing via the backdrop is disabled while submitting.
Camera Mode
Only available if the companion has the capture_visual_context tool/skill enabled. Toggling it on requests camera access and replaces the chat transcript with a live self-view video feed, with the assistant's current spoken response scrolling underneath as word-timed subtitles. An X button in the corner exits back to the transcript. If the camera stream isn't available, it shows a "Camera unavailable" placeholder instead of the feed.
Session End Modal
Appears automatically the moment a session ends (timeout, manual end, or connection drop). If the companion has feedback collection enabled, it shows a 1–5 star rating and an optional text field; Submit saves both to the chat record. The session's ID is shown with a copy-to-clipboard button for reference. If feedback collection is off, only a Thanks dismiss button shows. Skip/dismiss never navigates you away - the ended session's transcript stays visible in place, replaced by a Start New Chat button that clears it and resets to the landing view.
States
- Loading - a skeleton layout (carousel placeholder, title bar, icon row, input bar) shown while the companion's data loads.
- Invite error - if the companion requires sign-in with a specific invited email and you're not signed in as that user (or the invite link itself is invalid/expired), you get a dedicated "Sign In Required" or "Link Invalid" screen instead of the chat UI.
- Companion unavailable - shown if the companion can't be found, was deleted, or (for owners specifically) is temporarily broken; includes a Retry button.
- Service unavailable banner - if the organization is out of credits or its subscription has ended, a banner appears above the transcript (for owners: a direct link to top up or renew; the companion and its chats are flagged for deletion on subscription end) - this doesn't block testing entirely if a session is already active, but blocks starting a new one.
- Session ended by time limit - if the companion has a session duration limit and it's hit, a persistent amber banner appears below the transcript once the session ends.
- Empty history drawer - "Your conversations will appear here" when you haven't tested this companion yet, or "No matching conversations" when a drawer search returns nothing.
Who Can See It
The /test route is owner-only - it requires you to be signed in as a member of the companion's organization. There is no public equivalent of this exact URL; visitors experience the same underlying chat component through the public/share page instead, with public-visitor restrictions applied (interaction gating, no owner-only info in the Share dialog, etc).
What to Check Before Publishing
Testing isn't a one-time thing - repeat it after every round of changes. Things worth deliberately checking each pass:
- Greeting - does it say the right thing and feel on-brand? See Greeting Message.
- Personality and tone - ask a range of questions and see if the voice stays consistent. Adjust Personality if needed.
- Knowledge and accuracy - ask about topics you've added to Knowledge or Resources and confirm the answers are accurate and sourced correctly.
- Voice - if audio output is enabled, listen for naturalness and pacing; adjust in the Voice tab.
- Skills and tools - trigger each enabled skill directly (web search, image/video display, camera mode if enabled) and try switching languages mid-conversation.
- Guides - run through each published guide from the info popup, including branching paths and any Action-block ticket creation.
- Quick Start Prompts - confirm they appear and each produces a sensible response.
- Data collection and feedback - if either is enabled, go through the intake form and the end-of-session rating at least once to confirm the fields match what you configured.
Remember: every one of these test runs is a real session - it counts toward your usage and shows up in your own Insights. Use it accordingly, but don't expect it to be scrubbed from your stats afterward.
Last updated on