How it works
How Quack learns your product and works your queue
A trainable agent is only as good as what it knows. Here is how Quack builds that knowledge and applies it across your incoming tickets.
Steps 1 and 2
Ingest and learn from what your team already knows
Quack reads your product documentation and your closed support ticket archive. Most teams have both. Most AI tools ignore the tickets entirely. We don't.
- Documentation formats supported: Markdown, HTML help centers, PDFs, plain-text files. Point Quack at a URL or upload files directly.
- Closed ticket parsing: Quack reads the question, the human agent's answer, and any customer follow-up to understand how your team talks about the product, not just what the docs say.
- Continuous re-training: every time a ticket is resolved, it becomes new training data. Quack gets more accurate as your team keeps working, without any manual effort on your end.
Learning is scoped to your data only. Quack does not pull from external web sources or shared AI training corpora. Its knowledge is your product, nothing else.
Step 3
Answer with your product's voice
When a new ticket arrives, Quack searches its product model for relevant documentation and resolved tickets, then composes an answer grounded in your actual content.
Quack uses a confidence threshold to decide whether to answer or escalate. You set the threshold. Above it: the customer gets a specific, accurate reply referencing real product details. Below it: the ticket goes to your queue with Quack's draft and the articles it considered, so your agent starts with context, not a blank screen.
Responses reference actual steps, feature names, and product flows from your documentation. They don't pattern-match on generic support phrasing. A question about exporting gets the real export path, not a redirect to the help center.
Answered by Quack in 0.9 seconds
Step 4
Escalate with full context
When confidence drops below threshold, or the customer explicitly asks for a human, Quack escalates. Not with a hand-wave. With everything the human agent needs to pick up mid-conversation.
Full conversation thread
The human agent sees everything Quack and the customer exchanged. No re-asking "can you describe the issue?" The context is already there.
Knowledge articles referenced
Quack attaches the documentation sections it considered. The human agent can see exactly what Quack looked at, and why it wasn't confident enough to answer.
Confidence reasoning note
A short note explains why the ticket was escalated: low confidence, customer request, or out-of-scope topic. Helps agents triage faster and flag knowledge gaps for retraining.
Ready to stop re-answering the same questions?
We onboard early-access teams personally. Tell us your stack and we will have Quack reading your docs within 48 hours.