General settings

Open an Assistant and select General.

Runtime selection

Select the provider connection and model for the Assistant.

Visitor model choice

Select the models that the chat window shows to visitors. The Assistant model is always available. Each additional model that you select becomes an option in the chat window.

If you select no additional model, the chat window shows no model control. All conversations use the Assistant model.

The chat window shows a model only if your Organization has a provider connection for that model. If you remove the provider connection, the chat window stops showing the model. Conversations continue with the Assistant model.

The list shows no credential yet next to a provider without a connection. The line below the list counts only the models that the chat window can show.

Preview shows the new list immediately. Select Publish to give it to the widget, which uses the last published version.

Your own AI subscription has priority in Preview. If you set a default model in Settings, then AI, that model answers Preview turns and the chat window model does not apply.

Attachments

Set Let visitors attach files to give the chat window an attach control. It is off by default.

There are three ways to attach a file:

select + in the message box, then Attach a file
drag the file onto the message box
copy an image, then paste it into the message box

The chat window accepts these types:

PDF, Word and PowerPoint
Excel and CSV
text and Markdown
images (PNG, JPEG, GIF and WebP)

The maximum size is 8MB.

Old Office files (.doc, .xls and .ppt) are not supported. Save the file as .docx, .xlsx or .pptx.

Ciele reads the file into text and answers from that text. Ciele does not keep the file. The bytes are read one time and then discarded. There is no copy to download, to keep or to delete.

An image has no separate text extraction step. Ciele sends the image to your Organization's AI provider and asks it to read the image. This costs one AI call for each image. The result includes the text in the image and a short description of it.

An attachment stays available for the rest of the conversation. The chat window shows each attachment until the user removes it.

The Preview follows this setting as soon as you save. It shows the message box a visitor gets, before you publish. Teammate chats accept attachments at all times, because they have no visitor.

Select Publish after you change this setting.

Voice mode

Enable Voice mode to add microphone input and audio playback. It is off by default.

Connect a Google AI Studio, OpenAI or ElevenLabs API key in Settings, then AI.
In Voice input, select the transcription Model and Language. The default, Auto, detects the spoken language without translating the transcript.
In Voice output, select the playback Model, Language, and one of the three Voice choices. The default, Auto, follows the language of the latest input.
Select a voice's play button to hear a preview in the output language. Auto previews use English sample text.
Select Save changes. Publish the Assistant to update the widget.

The model lists use the connected providers' catalogues. Refresh reloads the lists. Voice choices follow the playback model. Model access also depends on provider permissions, billing and quota. The list shows three compatible voices, or fewer if the provider has fewer available. An existing selected voice stays in the list.

Input and output languages are independent and both default to Auto. For short recordings, select a known input language to improve recognition. You can also correct the transcript before sending. An explicit output language guides new AI-generated replies and their audio playback. Auto replies follow the latest message, including an edited transcript. Configured Flow messages and saved answers keep their original words. Audio playback reads that text without translating it.

Select the microphone beside Send to record a question. Recording stops after 60 seconds and accepts up to 4 MB. Select the stop control to transcribe the recording, or Cancel recording to discard it. The text appears in the message box. You can edit it before you select Send.

Select Read response aloud beside a completed answer to hear it. Select the same control to pause playback. Audio playback supports answers up to 4,000 characters. Voices are AI-generated.

Your browser must allow microphone access. Embedded widgets also need the microphone permission in their iframe. Update older iframe snippets from Publish if microphone access is blocked. Your operating system must also allow the browser or host app to use the microphone. On macOS, open System Settings → Privacy & Security → Microphone and enable that app. If macOS asks you to restart the app, do so before recording again. Allowing the site alone does not override this system setting. Ciele sends recordings to the selected provider for transcription. Ciele does not store the recording.

Identity

Configure these fields:

Title identifies the Assistant in the console.
Nickname appears in the widget.
Description records the Assistant purpose.
Avatar supplies the Assistant image.

Conversation defaults

Welcome message appears at the start of a chat.
AI disclaimer adds the configured disclosure text.
Answering style supplies standing response instructions.
Simplified thinking converts runtime progress into shorter Visitor-facing steps.
Suggested questions controls generated follow-up suggestions.
Quick replies add configured starter actions.

Save and publish

Change one group of settings.
Save the draft.
Test the draft in Preview.
Publish the Assistant when the result is correct.

The live widget continues to use the previous Publication until step 4 completes.