Files and Attachments

Dawn treats files as part of the workspace context, not just one-off message blobs. Users can attach files, mention existing files, preview generated outputs, and keep conversation artifacts available after the original message scrolls away.

How Files Work In Dawn

Dawn uses files in three common ways:

• Uploaded files: files a user adds to the composer.
• Mentioned files: existing workspace or thread files referenced in a prompt.
• Generated files: files created by Dawn tools, scripts, image generation, conversion, transcription, or document work.

Files can appear as message attachments, in thread file lists, and in preview panels. The important rule is that the durable file should exist once, and attachments should point to it rather than creating unnecessary copies.

Uploading Files

Attach a file when the agent needs to inspect it for the current request.

Use uploads for:

• documents and exports you want summarized or transformed;
• screenshots or images you want analyzed;
• audio or video you want transcribed;
• CSV, JSON, log, or text files that need processing;
• reference material for a workflow, command, or script.

Large uploads can take time. If an upload fails because it is too large, split the file or use a smaller export where possible.

Mentioning Existing Files

When a file already exists in Dawn, mention it instead of uploading it again. File mentions are the right path when the agent should use an existing generated file, prior attachment, or workspace file.

Mentioning an existing file should reference that file. It should not create a duplicate copy just because the file is used in a new prompt.

Use file mentions when:

• you want Dawn to revise or summarize an earlier generated file;
• you want a script or tool to use a known input;
• you want a new thread to continue from a file produced elsewhere;
• you want to avoid re-uploading the same document.

Generated Files

Dawn can create files as part of agent work. Examples include:

• transcripts from Voice Intelligence;
• image outputs and thumbnails;
• converted images or documents;
• markdown reports;
• JSON or CSV outputs from JavaScript scripts;
• HTML artifacts;
• screenshots and document staging outputs.

Generated files should remain available after the response is complete. If the agent attaches a generated file to a message, the attachment should download or preview from the underlying durable file.

Previewing Files

Dawn shows previews where possible so users do not have to download every file just to inspect it.

• Image thumbnails should load quickly in lists and message attachments.
• Full-size previews should open when you explicitly preview or open a lightbox.
• Text and document outputs can be opened from the file surface when supported.
• Attachments that cannot be previewed should still be downloadable.

If a thumbnail looks stale or missing, open the full preview or refresh the thread. If the full file fails to load, the file may not have been persisted correctly or the current user may not have access.

Thread Files

Thread file surfaces help collect files related to the active conversation. They are useful when the message history is long and you need to find an output without scrolling back through the full thread.

Thread files can include uploaded files, generated files, and files attached to assistant responses. The list should not show internal preview files as separate user-facing files.

Seeing What Dawn Is Doing With Files

File-heavy work can take several steps. Dawn can expose more detail about what an agent is doing while a step is running, including the inputs it is operating on when verbose progress is enabled.

Details stay collapsed by default so normal chat stays readable, and they expand when someone wants to see exactly what Dawn is working with. That is useful for debugging and trust: if Dawn is converting a file, opening an attachment, cropping an image, or generating an output, you can inspect the work without turning every run into a noisy log.

Good File Prompts

Be explicit about whether you want Dawn to inspect a file, transform it, or create a new output.

Use the attached CSV and create a new markdown report grouped by customer segment.

Open the mentioned transcript and extract action items into a CSV file.

Preview this image and generate three smaller variants for the blog post.

Troubleshooting

• The wrong file was used: mention the exact file or attach only the intended input.
• A generated attachment returns 404: refresh the thread and check the thread files panel. If it is still missing, ask Dawn to recreate the output.
• A file preview is slow: use the thumbnail/list preview first, then open full size only when needed.
• A file appears twice: use the existing file mention rather than re-uploading it.
• A file is unavailable in another workspace: files are scoped by workspace and access policy.

List the files available in this thread and explain which are uploads, mentions, or generated outputs.

Create a new generated file from the mentioned input without duplicating the original file.