AI assistant
Train the AI on your business documents — product catalogues, FAQs, price lists, and websites — so it answers customer questions instantly and accurately.
How the AI works
HopeChat AI learns from the documents you upload: PDFs, text files, images, and website URLs. When a customer messages you, the AI searches your knowledge base for the best answer and replies automatically. If it's not confident or the question is outside your knowledge, it escalates to a human agent so the conversation never goes cold.
Set up the AI
- 1
Open the AI page
Go to AI in the sidebar. The page shows your current knowledge base and configuration.
- 2
Upload documents
Click Upload and add your business files — price lists, FAQs, product descriptions, policy documents. Each file is indexed and searchable by the AI.
- 3
Add website URLs
Paste any public website URL you want the AI to learn from. HopeChat crawls the page content and adds it to the knowledge base automatically.
- 4
Configure behaviour
In the AI Config section you can set the system prompt (how the AI behaves), decide whether it escalates unknown questions, and review the training data to remove outdated or incorrect entries.
When the AI replies
- Confident answer — the AI sends a reply based on your documents, with a brief source attribution.
- Escalation— when the AI can't find a good answer, or the customer uses one of your trigger words, it lets the customer know a human will follow up and flags the conversation in your inbox.
- Human handoff— the conversation appears in your shared inbox wearing the reserved escalation tag (default "Needs human") so agents spot it at a glance. If auto-pause is on, the AI goes quiet until someone takes over.
Escalation to a human
Escalation is off by default. In the AI Hub under Configuration → Escalation to a human, switch on "Escalate when a human is needed" to decide when the AI hands a conversation to your team.
- Trigger keywords— any message that contains one of your words (matched whole, ignoring case) escalates the chat. The defaults cover things like "human", "agent", "refund", and "complaint".
- Auto-pause the AI — when on, the assistant stops replying as soon as a chat is escalated, so it never talks over your agent.
- Contact tag— escalated contacts are labelled with a reserved system tag (default "Needs human") that your team can filter on but can't delete.
- Needs attention filter — use it in the inbox to jump straight to every conversation waiting for a person.
From the conversation itself an agent can Take over (pause the AI and reply manually), Return to AI (resume automated replies), or Resolve the escalation once the customer is helped. Once an agent handles the chat the reserved tag is removed automatically — so it never lingers after the handoff is dealt with.
Control when the AI replies
By default the AI replies at all hours. In the AI Hub under Configuration → AI availabilityyou can switch on "Only reply during set hours", pick your time zone, and set a start and end time for each day. Days you turn off are treated as closed.
- Inside the window the AI answers incoming messages exactly as usual.
- Outside the window the AI stays silent and leaves the conversation for your team — no auto-reply is sent and no credit is used.
- Overnight windows (for example 22:00 to 06:00) are supported and wrap past midnight.
Best practices
- Keep documents current — update your knowledge base when prices, policies, or products change.
- Start with your FAQ — the questions customers already ask are the highest-value content for the AI.
- Review escalated conversations — they often reveal gaps in your knowledge base that you can fix by adding a new document.