Files, data, and generated outputs
Bergur DavidsenUpdated 2026-07-14
Usable Chat can work with images, documents, structured data, generated files, images, applets, and canvas output. The correct path depends on whether the material is temporary conversation input, reusable structured data, or an output that must be retained elsewhere.
File and data features depend on the selected model, tools, account, and embed configuration.
Attach a file or image
Use the composer attachment control, drag and drop, or paste an image where supported. Review the staged attachment before sending.
Typical inputs include:
- images for visual questions;
- PDF and text documents;
- source code and configuration files;
- CSV or spreadsheet data;
- other files allowed by the current uploader.
An attachment is part of the prompt context. It does not automatically become a durable Usable workspace asset or a memory fragment.
Do not upload secrets or regulated data unless the selected provider route, workspace boundary, and organizational policy permit it.
Choose Stage or upload for structured data
For CSV and similar structured data, released Chat can offer two paths:
- Stage for Chat: put the dataset into the current Chat data session for querying and analysis.
- Upload to workspace: store the file as a durable workspace resource when supported and authorized.
Choose Stage for temporary exploration. Choose Upload when the file should be discoverable and reusable under workspace permissions.
Staging may infer column types and provide a preview. Inspect the preview before sending. Type inference can misread identifiers, dates, decimal formats, or columns with mixed values.
Analyze data
Where enabled, Chat can use data tools and a sandbox to inspect staged datasets, run code, create charts, or generate derived files.
Ask for explicit checks:
Inspect the staged dataset, report row counts and missing values, state any type assumptions, then calculate the requested totals. Do not modify or export data until I approve the method.Generated analysis is only as reliable as the source data and code. Review formulas, filters, units, and sample rows before using the result.
Sandbox and code execution
Code execution runs in a controlled environment with limits on time, memory, persistence, and available packages. It is useful for calculations and file transformations, not as a replacement for your production runtime.
Treat files produced in the sandbox as temporary until you download or copy them to an approved durable location. Never assume a later conversation can access an earlier sandbox path.
Do not run untrusted code simply because it appeared in an attachment. Ask Chat to inspect it first, and avoid network, credential, and destructive operations.
Generated files and images
When compatible tools are enabled, Chat can generate downloadable files or images. The conversation may show a file card, preview, or link.
After generation:
- Open or download the output.
- Verify its contents and file type.
- Scan or review it according to your security process.
- Move the approved result to its durable system of record.
- Record sources and assumptions if the output supports a decision.
A generated filename, chart, or polished image is not evidence that the underlying content is correct or licensed for your use.
Applets, artifacts, and canvas
Chat can render richer interactive output where enabled. Released surfaces may describe these as applets, artifacts, or canvas actions.
Use them to explore or present information, but keep the source of truth in maintained content or files. Interactive state can be session-specific, and host applications may render or persist it differently.
For data-heavy work, preserve the input, transformation method, and final export separately from the visual surface.
Embed restrictions
An embedded assistant can hide or lock upload and workspace-file capabilities. The host can also provide parent-local tools for selecting or generating files.
The embed must not accept file behavior solely because the iframe requests it. The host and server must enforce user authorization, file size and type rules, scanning, and data retention policy.
Troubleshooting
The uploader is missing
Uploads may be disabled by the current Expert, model, account, or embed feature configuration.
Chat cannot read an image
Confirm the selected model and route support image input. Retry with a common image format and clear question.
A CSV column has the wrong type
Review the staging preview, correct the source format, or tell Chat the intended schema before analysis.
A generated file no longer opens
Sandbox and generated URLs can be temporary. Download important output promptly and store it in an approved durable system.
The result is too large
Ask Chat to summarize, filter, or generate a file instead of placing every row in the conversation.