How asking questions works
Q&A grounded in your document with AI retrieval and streaming replies.
Ask questions uses your file as context, not the open web. Answers stream in the right-hand panel while you read or listen.
Grounded answers
When you send a question, Speechdash:
- Retrieves relevant passages from the active document version
- Sends them with your question to the AI model
- Streams an assistant reply grounded in that content
The assistant does not invent facts from outside your document unless the text itself contains them.
Empty state prompts
With no messages yet, the panel suggests actions like:
- Summarize a section
- Explain a concept from the document
- Find specific information
Example questions
| Goal | Example prompt |
|---|---|
| Summary | "Summarize the introduction in three bullet points." |
| Clarify | "Explain the main argument in simpler terms." |
| Locate | "Where does the author discuss pricing?" |
| Compare | "How does section 2 differ from section 1?" |
| Study | "Give me quiz questions based on this chapter." |
Voice and avatars
Assistant messages show your voice-character avatar (same preset family as playback). Replies can auto-speak when streaming completes. See Assistant speak-aloud.
Credits
Each reply uses tokens plus a 1 credit flat surcharge per message. Long threads cost more on follow-up questions because context grows. Use Clear history to reset context and save credits.
Context sent to the model
On each reply, Speechdash includes up to 20 prior messages (user and assistant turns combined). Older messages stay visible in the panel but are not sent to the model. Clear history when a thread gets long or you start a new topic.
Persisted history
Ask questions history is saved per file across sessions. Clearing history removes stored questions and replies, not your file text.