Asking Questions
Once a knowledge base is selected (see Starting a Chat), the input bar is active and you can ask anything.
The input box
- Multi-line text field. You can paste long questions; the server-side validation accepts up to 50,000 characters per message, so practical length is rarely a problem.
- Placeholder: Ask Genie.
- Press Enter to send.
- Press Shift+Enter to insert a newline without sending.
- Pasting works normally. IME composition (Japanese, Chinese, etc.) is respected — Enter during composition won't accidentally send.
How to write a good question
Genie answers from your KB's documents. The clearer and more specific your question, the better the retrieval is. Tips:
| Style | Example |
|---|---|
| Specific | "What is the receiving-dock SOP for damaged pallets at the Madrid hub?" |
| Verbatim | If you remember a phrase from a document, include it: "What does the SOP say about the yellow tag procedure?" |
| Scoped | If your KB covers many sites, name the site or region |
| One thing at a time | Compound questions ("What is X and how does Y?") sometimes get partial answers — try splitting |
Avoid:
- Yes/no questions about hidden facts. "Is this allowed?" works better as "Under what conditions is X allowed?"
- Questions about things outside the KB. Genie will respond honestly that it doesn't know rather than guess.
- Personal information. Don't put names, employee IDs, or other PII into your questions — chat history is kept for audit purposes.
Multi-turn conversations
After Genie answers, the input box reactivates so you can ask a follow-up. Past messages stay on screen and Genie sees them too — this means follow-ups can use pronouns and shorthand:
You: "What's the cancellation policy for VIP customers?"
Genie: "VIP cancellations are free up to 24 hours before the appointment, after which a 50% fee applies. [policy.pdf#page=12]"
You: "And for standard customers?"
Genie: "Standard customers can cancel free up to 48 hours before; after that the full fee applies. [policy.pdf#page=8]"
The follow-up "And for standard customers?" worked because Genie kept the conversation context.
Streaming responses
After you press Enter, the answer starts arriving word by word with a small typewriter effect. This is normal — Genie streams the response so you can start reading immediately rather than waiting for the whole answer.
Genie currently has no way to interrupt a streaming answer. If you sent a question by accident, wait for the response to finish (usually 5–20 seconds), then start a new chat. The wasted answer doesn't cost you anything.
While Genie is generating, you see Generating answer followed by three animated dots and a spinner, until the first word arrives.
What if there's an error?
A red error block appears with the error message and a Retry button that resends the same question.
Common causes:
| Error | Likely cause |
|---|---|
| Generic "request failed" | Network blip — usually Retry works |
| "Your message contains content flagged by the OpenAI content filter" | The question triggered safety filters — rephrase |
| "Your message exceeded the context length limit" | The full conversation (current + history) is too long for the model. Start a New Chat. |
| Long pause then silence | Token may have expired — refresh the page and re-sign-in |
See Error Messages for the full list.
Conversation length
Within a single chat session, every previous message you and Genie have exchanged is sent back to the model with each follow-up. This is what makes pronouns and shorthand work, but it also means the prompt grows with every turn.
Because the model has a maximum context window, very long conversations can eventually hit the "Your message exceeded the context length limit" error described above. When that happens, start a new chat for a fresh topic. Each chat is saved separately and stays accessible from Chat History.
There is no fixed message-count cap that trims older turns automatically — the cut-off is driven by the model's context window, not a number we manage.