AI Model · 0.8.1

AI attachments, conversation history and interrupted requests

Attach requirements or a diagram, keep track of a proposal and handle an interrupted reply without blind retries.

On this page

Use Attach files when a diagram or a short specification is easier to provide than a long message. Attach only material needed for the model you want to create.

Supported attachments

File Limit
All selected files Up to 5 files and 10 MB combined
TXT, Markdown, JSON, YAML Up to 128 KB per file
PDF Up to 3 MB per file
PNG or JPEG Up to 8192 pixels per side and 20 megapixels

The interface uses MB/KB labels; limits are based on 1024-byte units. Remove an attachment with its chip's delete action before sending. An attachment is input to the request: it does not become a project file or an uploaded asset in the generated application.

Do not attach passwords, API keys, access tokens, private database dumps or customer records. A schema with fictional examples is usually enough. Your prompt, necessary model context and selected files are sent to the AI service and its model provider for processing; local chat storage does not make the request local-only.

What the local history keeps

Conversation text is saved locally for the current project. Attachment names may remain in a message, but their contents are not saved in the chat. Attach a file again if a later request needs to read it. Only bounded recent context is sent with each request; the visible history is not unlimited model memory.

Use the trash icon in the conversation header to open Clear local chat?, then confirm Clear chat. This removes the local conversation. It does not remove accepted entities or apply/discard the pending proposal. Unapplied proposals, selected files and unfinished input are not a durable workspace: leaving the page warns you that they will be discarded.

Cancel, wait or retry?

Cancel request stops waiting in the client. It does not guarantee that the provider stopped processing or used no tokens. A request can finish remotely even when the reply does not reach your application.

If the connection fails or the waiting time expires:

  1. Keep the error message and inspect Account / AI usage.
  2. Use Refresh access to get current quota information.
  3. Check whether a proposal or model change is already present before repeating an action.
  4. Submit a new request only when you understand the previous outcome. Requests are not automatically retried.

Clearing history, discarding a proposal or undoing a model change does not reverse provider usage. If the reply is repeatedly rejected, request a smaller set of new entities and check the creation-only rules.