Skip to content

Using the Chat

Thorsten Sommer edited this page Nov 6, 2026 · 9 revisions

Deutsch: Den Chat nutzen

In the chat, you talk to the AI model of your choice. This page explains the chat page, sending messages, working with the answers, and the options of the chat.

The chat page

  • The title shows where the chat belongs: "Chat in Workspace" with the name of the workspace, or "Disappearing Chat" for a chat outside the workspaces.
  • The provider selection below the title decides which model answers. You can switch it between messages.
  • The input field shows the provider in use in its label. Below it, a counter shows how many tokens the chat uses.
  • The toolbar below the input field holds the buttons for new chats, templates, attachments, formatting, the system prompt, profiles, tools, and data sources.

At the right end of the toolbar, AI Studio reminds you: "LLMs can make mistakes. Check important information."

Writing and sending

Type your message and press Enter, or click the send arrow at the end of the input field. Press Shift+Enter for a new line. You can change this in the "Chat Options" under "Shortcut to send input":

  • "Enter is sending the input": the default.
  • "Modifier key + enter is sending the input": Enter alone starts a new line; Shift, Ctrl, Alt, or Cmd with Enter sends.
  • "No key is sending the input": only the send arrow sends.

The buttons "Bold", "Italic", "Heading", "Bulleted List", and "Code" format the selected text in Markdown.

To attach files, click the paperclip or drop the files anywhere on the chat page. See Attachments.

While the AI answers

The answer appears while the model writes it. Meanwhile, you can't change the input, the provider, the profile, the tools, or the data sources; your choices apply again from the next message.

  • "Stop generation" ends the answer early. The part already written stays.
  • You can open another chat or another page in the meantime. The answer keeps being written, and the "Chat" icon in the navigation bar changes its color while it does.

When the model uses tools, a box shows "Using tool:" and the name of the tool.

The token counter

The counter below the input field shows how many tokens the chat uses, for example, "approx. 12,000 of 128,000 tokens", including your current message and the share of tools. It turns orange when the chat uses 80 percent of what the model can handle, and red when it uses all of it or carries more images than the model accepts. Long chats then become slow, expensive, or fail; start a new chat, or remove messages you no longer need.

"approx." means that AI Studio estimates the number. It counts with a tokenizer of its own, which differs from that of the model by a few percent on prose and by more on code. Once the provider reports how many tokens the chat used, usually with the first answer, the number is exact and "approx." goes away. It comes back when you delete a message or switch to another model, and it stays with providers that report nothing. The share of the message you are writing ("plus approx. … for your message") and, while tools work, their share ("of which approx. … from tools") are always estimates. When the preview feature for data sources is on, a tokenizer file for the model in the expert settings of the provider makes the number exact from the start.

The system prompt

The eye in the toolbar, "Show the complete system prompt the AI receives", opens the system prompt of the chat: the instructions the AI receives ahead of your conversation. AI Studio assembles it from up to five parts and shows each under its own heading, in the order in which the AI reads them:

  • the current date and time;
  • the default prompt of AI Studio, the prompt of your chat template, which replaces it, or the prompt of an assistant;
  • data from your data sources;
  • your profile, unless the chat template doesn't allow profiles;
  • the instructions for your tools, worded for this chat.

A chat which you start from the result of an assistant, for example, with "Send to ..." and "New Chat", gets the prompt which the assistant chooses for it: the default prompt of AI Studio, the system prompt of the assistant, or a chat prompt which the assistant writes for your follow-up questions. A chat prompt keeps the role of the assistant and the settings you chose, such as the language, but not its rules for the result, such as a fixed format. The heading of this part names the assistant, and for a chat prompt, a note below it says that it differs from the instructions with which the assistant created its result. Requires AI Studio v26.11.1 or newer.

The data from your data sources is what AI Studio sent with your last message. With classic RAG, AI Studio searches your data sources again when you send your next message and replaces this part; with Semantic Search, the AI receives what it finds as the result of its search. Which passages an answer received, its sources show. You can copy each part on its own, or the whole prompt with "Copy the complete prompt". Requires AI Studio v26.11.1 or newer.

Working with messages

Each message has a header with "You" or "AI" and the time. Its buttons depend on the message:

Button Where What it does
Copy, "Copies the content to the clipboard" every message Copies the message as Markdown, with its sources.
"Edit" your last message Puts your message and its attachments back into the input field. If the AI has already answered, AI Studio asks first and removes the answer.
"Regenerate" the last answer Asks the current provider again. Useful after switching to another model.
"Roll back to this response" earlier answers Removes everything after this answer, including attachments.
"Removes this block" every message Removes the message after asking you.
"Export AI response" answers Saves the answer as a document, or a single table or code block; see Pandoc and File Formats.
"Send to ..." your messages and the answers Sends the text to an assistant, or into a "New Chat", where it waits in the input field.

Code blocks in a message have their own save button, for example, "Save as Python (.py)".

AI Studio colors the code in Python, JavaScript, TypeScript, Java, C, C++, C#, Go, Kotlin, Rust, SQL, shell scripts, PowerShell, Lua, CSS, HTML, XML, JSON, and YAML. When a code block names no language, AI Studio recognizes it from the code, if it's clear: it colors the block, and its save button suggests the matching extension, for example, "Save as C# (.cs)". Requires AI Studio v26.11.1 or newer. Older versions color more languages, such as Ruby or PHP, but save a code block without a language as a text file.

Formulas have buttons as well: point at one to copy its LaTeX code or to save it as a LaTeX file, and the export menu of an answer offers its formulas, too. When a model puts a single formula into a code block marked as LaTeX, the chat shows it as a formula. Requires AI Studio v26.11.1 or newer. Older versions show formulas without buttons, and such a code block as code.

Tools and sources in answers

When the AI used tools, a wrench button next to the time shows them, for example, "Show 2 tool calls". Click it to see each call with its arguments and its result, and whether it was "Executed", "Failed", or "Blocked".

When an answer relies on sources, such as web pages or your data sources, they're listed below the answer with numbers like [1]. "Number of sources" in the header jumps to the list. See Tools in the Chat for setting up the tools.

New chats and copies

Button What it does
"Start temporary chat" Starts a new chat outside the workspaces.
"Start new chat in workspace '…'" Starts a new chat in the workspace of the open chat.
"Delete this chat & start a new one." Deletes the open chat after asking you, and starts a new one.
"Move the chat to a workspace, or to another if it is already in one." Moves the chat into a workspace you choose.
"Copy this chat & continue in the copy." Copies the chat under a new name and opens the copy.
"Save chat" Saves the chat, if you chose to store chats manually.

A new chat takes its name from the first ten words of your first message. You rename it in the workspace list; see Workspaces and Chat Search.

Chat Options

The gear icon next to the title, "Show the chat options", opens the "Chat Options". Changes apply at once. Besides the send shortcut and the preselection of provider, profile, and chat template, you find:

  • "Show the latest message after loading?": whether an opened chat shows its newest or its oldest message.
  • "Read changed attachments again?": whether AI Studio sends the new content of an attached file that changed since you sent it. By default, the AI keeps receiving the content the file had when you first sent it.
  • "Provider selection when creating new chats": whether a new chat starts with the provider you used last or with your preselected one.
  • "Provider selection when loading a chat and sending assistant results to chat": whether an opened chat continues with its own provider, the one you used last, or your preselected one.
  • "Default tools for chat": the tools every new chat starts with.

When something goes wrong

Errors appear as a notice at the edge of the window. The answer that failed is removed, and your message stays, so you can resend it with "Edit". Common causes:

  • "The API key might be invalid": check the key in the settings of the provider.
  • "The provider is overloaded" or "too many requests were sent": wait a moment and try again.
  • "The data of the chat, including all file attachments, is probably too large": start a new chat, or attach fewer or smaller files.
  • "The selected model is no longer available": choose another model in the settings of the provider.

When the input field reads "Select a provider first", choose a provider. When the provider selection says that no provider is configured or none meets the requirements, add a suitable provider in the settings; see Providers and Models.

Where to go next

Clone this wiki locally