Upload & Chat — File Limits & How Large Documents Are Handled
7 min read
Last updated: June 2026 · Applies to Unique AI Chat (production)
At a glance
Users can upload documents directly into a chat and ask questions about them. Files up to 512 MB are supported. As soon as a file is attached, it is included in the AI's context automatically for every answer in that conversation. The only way to take a file out of the conversation is to open the Uploaded Files panel and uncheck it (if every file is unchecked, no uploaded file is sent to the AI). The user does not need to instruct the agent in the prompt, and the agent does not get to decide whether to consult an included file. For large documents, the assistant retrieves the most relevant sections rather than loading the entire file into the model at once.
What is Upload & Chat?
Upload & Chat lets users attach files to a conversation in Unique AI Chat. After upload, the file is processed in the background so the assistant can search it and answer questions with cited sources.
Typical use cases:
Ask questions about a contract, report, or presentation
Summarise an uploaded document
Compare information across multiple uploaded files in the same chat
How are uploaded files used by the AI?
When a user uploads a file to a chat, using it is the default — the agent does not get to decide whether to consult included files. Every question in the conversation automatically triggers a retrieval step against the currently-included uploaded files:
Included by default. Every newly uploaded file is automatically in scope for the AI from the next turn onward.
The only opt-out is the Uploaded Files panel. Open the file picker in the chat input (tabs: Internal Files │ Uploaded Files), and uncheck individual files to exclude them. If you uncheck every uploaded file, none will be sent to the AI for that turn. There is no global "do not use my files" toggle outside this panel.
Not a prompt-engineering question. The user does not have to write "use my file" or "look at the document"; if the file's checkbox is ticked, it's already in scope.
Not an agent decision. For each turn, the orchestrator forces the retrieval tool to run against the currently-checked uploaded files; the LLM does not get to skip them.
Multiple files = one combined scope. All currently-checked files are queried together; the user does not need to pick which one to use, only which ones to leave out.
PDF “full-file” mode (experimental). On environments where the experimental full-PDF attachment mode is enabled, each checked PDF is attached to the model's input directly from the first message instead of being chunk-searched. Unchecking still excludes it. From the user's perspective the behaviour is the same: checked files are always used, automatically.
Files expire after a retention window (compliance). Uploaded files are automatically removed after an admin-configured retention period (typically 2 weeks in production). Once a file expires, the chat shows a warning that the AI can no longer consult it. To keep asking questions about it, the user must re-upload the file.
Talking point for customers: "Once a document is attached and ticked in the Uploaded Files panel, Unique AI is required to consult it for every reply — it is never a question of whether the assistant feels the file is relevant. To take a file back out of the conversation, untick it in the panel."
File size and upload limits
Limit | Value (production) | What happens if exceeded |
|---|---|---|
Maximum file size | 512 MB per file | Upload is blocked with: "File too large. The current limit is 512 MB per file." |
Files per upload | 30 files at once | Upload is blocked with a "too many files" message |
Note: An older 100 MB limit is no longer in effect. If colleagues or customers refer to 100 MB, the current production limit is 512 MB. Limits can be adjusted per environment by your Unique administrator.
Supported file types
By default, the following formats are supported in chat:
Documents: PDF, DOCX, XLSX, PPTX, CSV, HTML, Markdown (.md), plain text (.txt)
Images (when enabled on the Space): PNG, JPG, JPEG
Some Spaces restrict file types further (for example, Translation Spaces may only accept DOCX, XLSX, and PPTX). Upload is enabled by default on for Space setup.
What happens after upload?
Upload — The file is stored securely.
Processing — The system reads the document, splits it into smaller searchable sections ("chunks"), and indexes them. A spinner appears on the file badge while this runs.
Ready and ticked by default — When processing completes, the file appears in the Uploaded Files panel with its checkbox ticked. You can untick it there to exclude it from future answers, or leave it ticked (the default) to keep it in scope.
Forced retrieval at every turn for ticked files — From this moment on, every question you send in this chat automatically runs a retrieval step against the currently-ticked uploaded files. The agent cannot skip them; the user does not need to opt in. Unique AI then builds an answer with source citations.
Large documents and the AI context window
A common question: "Can the AI read my entire 200-page PDF in one go?"
Short answer: It depends — and that is by design.
Language models have a fixed "context window" (a maximum amount of text they can process in a single request). Unique AI handles large files using retrieval, the whole file fits into the LLM if the context window is large enough. The user does not know if only part of the file is seen by the Agent.
How it works for Q&A
Your document is split into many small sections during processing.
When you ask a question, the system automatically searches those sections — this step is forced for ticked uploaded files, not optional.
Only the best-matching sections are sent to the AI (typically up to ~40% of the model's input capacity for search results). The sections are ordere according to the original document when presented to the Agent.
The AI answers based on those sections and cites its sources.
What this means for users
Scenario | What to expect |
|---|---|
Specific question about a large file | Works well — the system finds the relevant sections |
Detail buried deep in a 500-page file | May be missed if the search does not surface that section — ask a more specific question or reference a section/page |
What users see when something goes wrong
Situation | User experience |
|---|---|
File too large | Error toast before upload starts |
Unsupported file type | "Invalid File Type" message |
Processing failed (e.g. corrupted file, malware, empty document) | Modal: "Unable to upload the following files" with details |
Spreadsheet too large (tables) | Processing may fail if row/column limits are exceeded (default: 5,000 rows × 100 columns) |
FAQ — quick answers for colleagues
Q: What is the upload size limit?
A: 512 MB per file, up to 30 files per upload batch (production default).
Q: Is the limit still 100 MB?
A: No. 100 MB was an older limit. Production is 512 MB.
Q: Does the AI read the whole file?
A: Depends on the context window setup by the admin and the file size.
Q: Does the assistant choose whether to look at my uploaded file?
A: No. Every file that is ticked in the Uploaded Files panel is automatically forced into the retrieval step for every question in that chat. The agent does not opt out, and the user does not need to mention the file in their prompt.
Q: How do I tell the AI to ignore an uploaded file?
A: Open the Uploaded Files tab of the file picker in the chat input and untick the file's checkbox. While unticked, it is excluded from the AI's context. Tick it again to bring it back in.
Q: What if I untick every uploaded file?
A: Then no uploaded file is sent to the AI on that turn — the assistant will answer from chat history and any internal-knowledge sources you have selected, but not from your uploads.
Q: Do I have to tell the AI "please use the file I uploaded"?
A: No. Attaching the file (and leaving its checkbox ticked) is enough — retrieval is triggered automatically on every turn.
Q: Can users choose which uploaded files to include?
A: Yes — via the Uploaded Files panel's per-file checkboxes (and the "Select all" toggle). Within the ticked set, retrieval is still automatic; the agent doesn't choose.
Q: Can limits differ per customer?
A: Yes. File size, file count, context window size and allowed types can be configured per environment by Unique operators.
Q: Can uploaded files be seen by other users?
A: No. Uploaded files are scoped to a single chat and are not visible to other users. If you start a new chat, even with the same agent, the uploaded files are not available. Files also expire automatically after a configurable retention period.