How do image and file uploads work, and what are the limits?

You can attach images and documents to a question, and the council will read them along with your text. Everything is prepared in your browser before it is sent, so payloads stay small and unsupported files are rejected early. This page lists what is allowed, the exact limits, and what to check on your models first.

The short version

Images

What is accepted

JPEG, PNG, WebP and GIF, up to 10 MB each, with up to 4 images per question. Images with more than 50 million pixels are refused, which mostly catches very large scans and panoramas. Any other format, such as HEIC, TIFF, BMP or SVG, is rejected. Convert it to JPEG or PNG first.

What happens to your image

Your image is never sent as-is. In your browser it is:

1. Scaled down so the longest side is at most 1024 pixels. Smaller images are not enlarged.
2. Placed on a white background, so transparent areas of a PNG or WebP turn white.
3. Re-encoded as a JPEG at 85% quality.
4. Checked against a size cap of roughly 1 million characters once encoded. If it is still too big, it is retried with lower quality and a smaller size, in up to three further steps, down to 512 pixels at 60% quality. If it still does not fit, you get an "image is too large after processing" error.

What this means in practice:

• Text in screenshots can get hard to read. At 1024 pixels, a full-screen capture of a dense page loses small print. Crop to the part that matters before uploading.
• Animated GIFs become a single still picture. The models do not see the animation.
• Colours on transparent images change. Light text on a transparent background can disappear against white. Export with a solid background if that matters.
• Photos taken on a phone are fine. They are downscaled automatically.

Files

What is accepted

Up to 5 files per question, up to 10 MB each. Supported types:

• Documents: PDF, DOCX
• Spreadsheets: XLSX, XLS
• Text and data: TXT, MD, MARKDOWN, CSV, TSV, JSON, JSONL, XML, YAML, YML, TOML, INI, LOG
• Web and code: HTML, CSS, JS, JSX, TS, TSX, PY, JAVA, C, H, CPP, CS, GO, RS, RB, PHP, SH, SQL

Older Word .doc files and PowerPoint files are not supported. Save them as DOCX, PDF or paste the text.

How files are read

Files are read in your browser and only their text is sent to the models, inside the prompt. Each file is wrapped with its name so the models can tell them apart. Because it is plain text, this works with every provider, including ones that cannot read images. The PDF, Word and Excel readers load only when you attach one of those types, so they do not slow down the app otherwise.

• PDF: text is taken page by page and labelled with page numbers. Reading stops after 100 pages, or sooner once about 20,000 characters have been collected.
• DOCX: the raw text of the document. Images, charts and layout are not carried over.
• XLSX and XLS: every sheet is turned into CSV text, labelled with the sheet name. Reading stops once the 20,000-character mark is passed. Charts and formatting are not included.
• Text and code: read as-is. A file that contains binary data is rejected.

The 20,000-character limit

Each file contributes at most 20,000 characters to the prompt, which is roughly 3,000 to 4,000 words or about 5,000 tokens. Anything beyond that is cut off and the models are told the file was truncated, so they know they are not seeing the end. With 5 files attached, that is up to about 100,000 characters in one question.

If your document is longer:

• Attach only the relevant chapters or sections as separate files.
• Split a long spreadsheet by sheet or by date range.
• Paste the key passage into the question instead.
• Ask about one part at a time and move on to the next.

Files that will not work

• Scanned PDFs and PDFs that are only pictures. There is no selectable text to extract, and the app does not do OCR. Upload the pages as images instead, to a model that can read them, or run OCR first.
• Password-protected or corrupted files. You will see a message that the file could not be read.
• Empty files, or files with no text in them.
• Files over 10 MB, even when the text inside would be small. Compress or split them.

Check your models before you upload

For images: every seat must be able to read them

This is the most important check. A model that cannot read images does not stop the council, but that seat answers from your text alone and never sees the picture, so it may guess or say it cannot see anything. The Judge then merges answers where one or more seats were blind.

Before uploading an image:

1. Open Model settings and look at each of the four seats. If a provider cannot read images, its card shows a note saying that seat answers from text only.
2. Check the model, not only the provider. Some providers host both image-capable and text-only models. Use Load live models to see what is available, and pick a vision model.
3. If you are not sure, run a quick test: upload a simple picture and ask "what is in this image?". A seat that cannot see it will say so.

Good free choices for image work are Gemini's free Flash, Gemma 4 31B and Nemotron 3.5 Super VL on OpenRouter. Cerebras is text-only, so avoid it for image questions. See What is the best combo of free models for the council? for a ready-made images lineup.

For files: check context length and input limits

Any model can read extracted text, but a small input limit can still break a seat.

• Short context windows. Cerebras's free tier has about 8K of context, which is not enough for even one full-length file plus the question. GitHub Models also has tight input limits. Keep these off the seats that will receive big attachments, or attach less.
• The Judge reads the most. It sees the three Thinkers' answers on top of the question, so give it the longest-context model you have.
• Rate and token limits. Five large files in one question use far more of a free daily allowance than a short question does. If a seat starts failing after several uploads, it has probably hit a limit, so switch it to another provider.

Specialties help

Pick the specialty that matches the file. A coding specialty fits source files and logs, and a research or reasoning one fits reports and spreadsheets.

Privacy

Files and images are processed in your browser, but their content is then sent to the providers you chose, as part of the question. Some free tiers, including Gemini's, may use prompts to improve their products. Do not upload contracts, medical records, financial statements, customer data or anything you would not paste into a public form. For sensitive material, remove the private parts first or use a provider whose terms you trust. If a file contains keys or passwords, delete them before uploading.

Quick checklist

• Up to 4 images and 5 files, each under 10 MB.
• Images: JPEG, PNG, WebP or GIF, and crop to what matters.
• Files: pick a supported type, and use text-based PDFs, not scans.
• Long file? Expect it to be cut at 20,000 characters, so split it.
• Every seat on a vision model if you attach images.
• Judge on a long-context model.
• Nothing private in the upload.

Troubleshooting