Skip to main content

Images & attachments

Attach files directly to a conversation when the material is temporary and specific to the current task.

Good uses for attachments

  • screenshots of an error
  • one configuration file
  • a document to summarize
  • an image to analyze
  • a log export for one debugging session
  • a small group of files needed for comparison

Model capability matters

A model must support the relevant input type. An image-capable route can interpret supported images; a text-only route cannot gain vision simply because an image was attached. Starfire can restrict model choices when an attachment requires a capability the selected model does not have.

Attachment vs Knowledge

Use an attachment when the file matters now. Use Knowledge when the file should be indexed and reusable across many conversations or projects.

Context budget

Extracted attachment content still consumes model context. Large files can be processed selectively, summarized, or retrieved rather than injected in full depending on the workflow.

Processing state

A file can upload successfully before all extraction or indexing work has finished. If the current workflow requires processed text or retrieval, wait for the relevant processing state.

Security

Do not attach secrets that are not necessary for the task. If a password, API key, or private key is accidentally exposed, rotate it rather than relying only on deleting the visible attachment.

Troubleshooting

If Starfire does not appear to use an attachment:
  1. Confirm the upload completed.
  2. Confirm the selected model supports the input type.
  3. Confirm the file format is supported by the current workflow.
  4. Ask explicitly for the model to inspect the attached material.
  5. For persistent sources, check Knowledge indexing instead of chat attachment state.

Files & context

Compare temporary attachments with Project and Knowledge context.