Build
Four things make an agent
Your content
Crawl your website, upload documents, add FAQs. Re-sync when they change.
Personality and scope
Tone, language, what it may and may not discuss, and when to escalate.
Flows and integrations
Collect details, look up an order, create a CRM lead, book a slot.
Channels and hours
Choose channels, and whether it runs always, after hours or in business hours.
Grounded
Answers from your knowledge, not guesses
The agent retrieves the most relevant passages from your knowledge base for every question and answers from them. If nothing relevant is found, it says so and offers a person instead of inventing an answer.
- Website crawler with safeguards against internal-network addresses
- Chunking and embeddings tuned for FAQs and policy pages
- Re-sync sources when your content changes
Handoff
A clean handover, every time
Handoff is not an error state. The agent passes the conversation to the right team with a summary, the details it already collected, and the reason it stopped.
- On low confidence, explicit request, or a topic you flag
- Respects business hours and team availability
- Customer is told what happens next
Safe to change
Test before customers see it
Save real conversations as test cases with the answer you expect. Run the suite after changing instructions or knowledge to catch answers that got worse.
- Simulator to chat with a draft agent
- Test suites with expected outcomes
- Versioned deployments to production
Where agents work
Deploy on the channels you choose
The same agent can answer on WhatsApp, your website widget, Instagram, Messenger, Telegram and in-app chat. On WhatsApp it respects the 24-hour customer service window.