Skip to main content
Find the symptom you’re hitting below and work through the fixes in order — they’re ordered from most to least common cause. Each one links to the page that covers it in depth. For shorter how-do-I questions, see the FAQ.
Most agent and answer issues trace back to one thing: a source that isn’t Completed. That’s always worth checking first.

​
My agent says it doesn’t know, or gives the wrong answer

The agent answers from the sources in its knowledge base. When it can’t find a match — or finds the wrong one — work through these in order.
Open Expertise → Knowledge Base and check the source’s Status. Only Completed sources are used in answers — content that’s still Unprocessed, Processing, or Failed is invisible to the agent. See Manage & reprocess sources.
If a relevant source is Completed but the agent still misses, the question wording probably doesn’t match your content. Add a Q&A pair phrased the way a real customer asks — Q&A is the most precise way to control an answer. Add a few variations if customers phrase it different ways.
If the answer is off-tone, off-topic, or guessing, refine your instructions — set the persona, scope, and a clear fallback (“if you can’t find the answer, say so — don’t guess”). Save each change as a new version so you can compare or roll back.
A stale or contradicting source can outrank the right one. Update or delete outdated sources — outdated content is the most common cause of wrong answers.
Verify the fix in the Sandbox, and check the answer cites the right source. To compare two full configurations, use the Playground.

​
A source is stuck Processing or Failed

Adding a source runs background processing — read, chunk, and index. Large or image-heavy files take longer.
Large or image-heavy files take longer — give it a few minutes; processing runs in the background even for big files. If it stays stuck or turns Failed, reprocess the source from its Actions menu (⋯).
Reprocess it first. If it keeps failing, the file may not extract cleanly — scanned PDFs (images of text) extract less reliably than digital PDFs. Try a text-based version, or paste the key content as Text or a Q&A pair. Supported file types are PDF, DOCX, and TXT — see Upload files.
The page may render its content with JavaScript, or hide it behind a login or cookie wall. Confirm the URL is public and reachable in a normal browser, then reprocess. If it still comes up empty, paste the key content as an uploaded file or Q&A. See Add a website.
The video likely has no captions, or it’s private/restricted. Use a public video with captions, or add the content another way. See Add YouTube.
Reprocess the source — it re-reads, re-chunks, and re-indexes the latest version. For a website or YouTube source it re-fetches the current content at the URL.

​
The chat bubble or widget doesn’t appear

Both snippets from the Embed/Install screen need to be in place, and your site’s cache cleared.
Confirm both snippets are present: the widget script in your <head> and the config (window.A2V2WidgetConfig = { chatbotId: "..." }) before </body>. Make sure chatbotId is filled in and that your site doesn’t strip <script> tags. Copy the snippet directly from the Embed/Install screen rather than retyping it.
Open your browser’s developer console on the page and look for errors. If your site enforces a strict Content-Security-Policy, allow the widget’s font and script sources.
Your site or CDN may be serving a cached copy of the page without the snippet. Hard-refresh the page, and clear any site/CDN cache after embedding.
The widget loads its own font and isolates its styles in a Shadow DOM, so it won’t clash with your site. If a strict Content-Security-Policy is blocking it, allow the widget’s font/script sources. See Embed & install.

​
My agent stopped answering / credits are exhausted

Credits are checked the moment an agent tries to act. At a zero balance, agents stop generating new answers — but nothing is deleted, and answering resumes as soon as credits are available.
Open Billing in your organization settings to check your balance. If it’s at zero, visitors see a notice that there aren’t enough credits to respond, and voice messages show “Insufficient credits to transcribe this recording.” See Credits explained.
Use Top Up Now to add credits immediately, or wait for your plan credits to refill at the start of the next billing period. See Auto-recharge.
Turn on auto-recharge so A2V2.ai tops up automatically when your balance gets low — set a Max recharges per month cap to bound your spend. For agents customers rely on, this is the difference between staying online and turning real visitors away.
A premium model costs more per message than a standard one, and high traffic multiplies that. Check which model your agent uses on Model & settings and compare costs in the Sandbox model selector. Document processing and voice transcription also draw credits. Track spending on the Usage page.
Use Refresh Usage on the Billing page to pull the latest figures. Top-up credits appear under Top-up Credits (“Remaining from recharges”), separate from your monthly plan credits. Some card payments need an extra verification step from your bank — the credits are added once that completes.
You may have reached your Max recharges per month cap, or a payment failed. If several payments fail in a row, A2V2.ai disables auto-recharge. Update your card in the billing portal, then re-enable it. See Auto-recharge troubleshooting.

​
A visitor can’t access a gated agent

A public agent can be open, restricted to your CRM, or behind a paywall. The fix depends on which mode it’s in.
On Visibility & access, a Private agent is hidden to everyone but you. Switch it to Public when it’s ready for real visitors.
That’s expected with the Restrict mode — only users from your CRM can use the agent, so visitors identify before the conversation starts and are captured as contacts. Switch to Use for Free if you want no-friction access. See Visibility & access.
On a Pay for Access agent, their access duration may have expired or their message credits may be used up. Check their status in the Subscriptions table. See Paid agents troubleshooting.
Pay for Access is finalized with the A2V2.ai team — after you enable it, they reach out to set the price and duration. If you haven’t heard back, contact support@a2v2.ai.

​
I can’t do something / a button is disabled

Some actions are limited by your role or your plan.
Billing changes — plans, payment method, and auto-recharge — are Owner-only. Ask your organization Owner. See Members & roles.
Only the Owner can invite or promote someone to Admin. As an Admin you can invite Users only. See Members & roles.
You’ve used every team seat on your plan. Add the Team Member add-on or upgrade your plan. Seat changes are handled by the Owner.
These are desktop features. On smaller screens you’ll be asked to switch to a larger display. Open them on a desktop or a wider browser window.

​
Processing or answers feel slow

Large or image-heavy files take longer to process — this runs in the background, so you don’t have to wait on the screen. Track status from the source list. If it stalls or Fails, reprocess it.
The agent streams its reply in real time — while it works, you may see it thinking and using tools before the final answer appears. Heavier premium models can take a little longer per answer than lighter ones; compare in the Model reference.

​
Still stuck?

If none of the above resolves it, email support@a2v2.ai and include your organization name and the agent you’re working on.

FAQ

Quick answers to common questions.

Manage & reprocess sources

Track source status and retrain.

Credits explained

How credits work and what happens at zero.

Embed & install

Add your agent to your site the right way.