Chatting
A chat keeps your messages, answers and related context together. Available models, tools and input options depend on your organisation, the selected model and any active agent.
Chat header and message box
Example instance with sample data. Available controls depend on your role and configuration.
Start a chat and choose a model
- Select New chat.
- Open Select model and choose a model by its described role, such as general work, image understanding or coding.
- Write your request, then select Send or press
Enter.
An agent may set the model, thinking, web search and tools. Locked settings cannot be changed for that chat. Start a new chat without the agent, or with a different agent, if needed.
A useful request states the task, context, audience and output format. Split long tasks into steps and check important results.
Control an answer
While an answer is being generated, select Stop to halt it. Text already produced remains visible. Depending on state, actions below an answer include:
- Copy copies the visible answer.
- Retry generates another answer to the same preceding message.
- Continue continues an incomplete answer when offered.
- Helpful and Not helpful submit a rating.
- Report problem opens a form where you can describe the issue and consent to sharing the listed content.
AI output can contain mistakes. Check figures, quotations, legal information and specialist advice in particular.
Edit your message
Hover over your message and select Edit. Change the text and select Save and send. The chat continues from that point using the edited message, so later answers may be replaced.
Use files and images
Select Attach file to choose an existing file or upload a new one. Wait for the upload to finish before sending.
- Supported documents can be used as knowledge sources.
- Images are sent only to models that support image understanding.
- Your organisation may set a maximum number of files or images per message. Intra AI warns you when an item will not be sent.
- Project files and knowledge supplied by an agent may also be considered.
If no relevant passage is found in your documents, Intra AI may say so in the answer. Review source cards and open the cited file before relying on a claim.
Use web search or deep research
When web search is enabled, choose one mode for a message:
- Search lets the model search the web while answering. Used web sources appear with the answer.
- Research performs a multi-step search, reads sources and writes a report. Progress includes stages such as planning, searching, reading and writing. You can stop the run and keep content produced so far.
The modes are mutually exclusive. An agent may set the available mode. Web sources can change or be wrong, so open and check decisive evidence.
Reasoning effort control for a supported model
Example instance with sample data. Available controls depend on your role and configuration.
Set thinking effort
Supported models show Think in the message toolbar. Open Reasoning effort to use the on/off switch or choose Low, Medium or High, where the model supports those controls. Greater effort can help with complex analysis, but usually takes longer and uses more thinking tokens. An agent may lock this setting.
An expandable thinking panel shows only what the application provides for that answer. It is not evidence that the result is correct.
Quote text and use Markdown
Select text in a message and choose Ask Intra AI to attach it as a quote to your next message. Use Remove quote to discard it.
Select Md to open Markdown tools. You can format headings, lists, quotations, tables and code, then switch between editing and preview. Your personal Markdown input preference is retained.
Use saved prompts
Enter / at the start of an empty message box. Choose a globally provided or personal prompt and search by its name or content if needed. Manage personal prompts under Settings → Prompt templates.
Review and confirm tool actions
When an agent or model uses a tool, the chat shows its status and result in a card. Open the card or inspector to review available input, output and technical details.
Actions that write to another system may require confirmation. Check recipients, destination, content and other fields carefully. Select Confirm, Cancel, or tell the assistant what to change.
Use a temporary chat
Select Turn on temporary chat in the chat header before starting the conversation. The ghost icon switches the mode and opens a new chat. The temporary chat is visibly marked.
Temporary chats do not appear in normal history and do not use your memory. They are not intended for projects. Copy anything you need before leaving or closing the chat, as you cannot reopen it from the sidebar afterwards.
Organise and manage chats
Open the menu beside the chat title:
- Rename changes the title.
- Archive removes the chat from the normal list; open Archived to restore it.
- Delete permanently removes the chat after confirmation.
- Leave chat appears for chats you participate in but do not own.
In the sidebar, you can Pin or unpin a chat. Under More actions, owners may be able to bookmark, share or add the chat to a project.
When the context becomes long
Intra AI shows how much of the chat context is used. In long conversations, earlier context may be summarised automatically. Messages remain visible; the assistant uses the condensed summary for later answers. When offered, you can select Summarise now and view the summary later.
Troubleshooting
| Problem | What to do |
|---|---|
| A tool is missing | Check the model and agent, then ask your administrator whether the feature is enabled. |
| A file is ignored | Wait for upload, check its status and source notice, then ask for the relevant passage more precisely. |
| An answer stops early | Select Continue or Retry. |
| Research takes too long | Open progress or select Stop research. |
| A write action is unclear | Cancel it and describe the required change in the message box. |
Related: Sharing chats · Exporting chats · Memory