AI chat guide: models, follow-ups, context and usage
Start an AI conversation, choose or switch models, attach useful sources and refine the answer. Find your history, context controls and request details.
Enlarge preview
A draft, a follow-up and a shorter reply
Start a chat and choose a model
Sign in, open New Chat and select Chat. Use the model picker to choose a model, or let Auto route the request. Type your question or paste your notes, then press Send.
Give the model the source it needs
Paste useful notes or attach a supported file. Name the source facts, the format you need and any missing details. Ask for a source location when it will help you check the answer.
Revise in the same conversation
Use the next message for a specific change. Name the facts or constraints that must remain. Compare the revised answer with the original source before using it. Switch models deliberately when needed; earlier conversation content can become context for the newly selected provider.
Find your completed request usage
Open a response’s More actions → Details where supported, or use Settings → Usage History and filter by date, model or activity. Match the request to its completed entry. Input, context, output and tools can affect usage.
Access and cost
| Decision | What to check |
|---|---|
| Free Chat | Finite signup credits; minimum chat cost and selected-model/tool rules apply. |
| Plus default Arya | Included in-app chat under the current conditions; other models and tools can cost credits. |
| API | Separate authenticated credit usage, including Arya. |
| Work and live Voice | Plus required; their applicable usage costs are separate from ordinary Chat. |
Temporary chats and saved context
Temporary Chat does not appear in ordinary history and does not create memories. Its default personalization ignores saved memory and collection instructions; check the setting before relying on that behavior. Temporary does not mean that no processing, security or billing records exist.
How provider processing works
Gab sends the relevant prompt, files and enabled context to the selected provider through its provider account. Personal details in those inputs may still be shared. The privacy guide explains processing and controls.
If a request cannot run
Read the displayed error before retrying. Check sign-in, model eligibility, available balance, attachment limits and completion status. Repeated sends can create additional requests. A stopped or failed response does not by itself prove a refund; inspect completed usage and your account records.
Keep the conversation moving.
Ask a question or paste your notes into Chat. Follow up with the change you want and the facts that should stay.
Open ChatTry a complete first prompt