Frequently Asked Questions

Every AI chatbot has a limited working memory — called a context window. When you upload a scanned PDF directly, the AI has to spend part of that memory deciphering garbled OCR text, broken formulas, and formatting junk.

Even when it can figure out what the text probably says, that deciphering process burns through tokens quickly. The result: shorter answers, earlier cutoffs, and more hallucinations — cases where the AI confidently makes things up because it couldn't read the source clearly.

Ellide cleans and compresses the document before it reaches the AI, so the model spends its context on actually answering your question — not on guessing what "Artlficlal lntelligence" means.

Ellide's cleanup quality depends on which AI model processes your document. Our default model is fast and handles most documents well, but heavily degraded scans or unusual formatting can still trip it up.

Higher-tier models produce dramatically better results — especially on messy scans, complex tables, and technical content with formulas. We're rolling these out with usage limits initially, and they may become a paid feature down the road.

If accuracy is critical for your use case, try reprocessing the document — results can vary between runs, and a second pass often catches what the first missed.

Ellide supports a wide range of document types:

• PDFs — scanned or digital, including multi-hundred-page documents
• Images — PNG, JPG, TIFF, and other common formats
• Presentations — PowerPoint (.pptx, .ppt)
• Word documents — .docx and .doc
• Plain text — .txt, .md, or just paste directly

File size limits: 50 MB without an account, 200 MB with a free account. These limits exist to prevent abuse and keep the service running smoothly for everyone.

Markdown is the right choice for most people. It's essentially AI's native language — structured enough to preserve headings, lists, and formatting, but lightweight enough to keep token counts low. If you're pasting output into a chat window, Markdown is almost always what you want.

JSON is better when your document has strict, tabular, or hierarchical structure that Markdown can't capture cleanly — think spreadsheets, form data, or structured records. Developers may also prefer JSON for automated workflows.

TL;DR: Pasting into a chat window to study or get answers? Markdown. Working with structured data or building something technical? JSON.

Ellide never saves the documents you upload. Your files are processed in memory and discarded immediately after.

If you have an account, you can optionally save the cleaned output for later access — but the original document is never stored.

All AI processing runs on the enterprise tier of Google Cloud's Vertex AI, which does not use your data for training and only retains data in cases of policy abuse. You can read the full details in Vertex AI's data governance policies and Google's content safety policies.

Not through the web app — and honestly, you shouldn't be pasting API keys into any website you don't fully control.

If you'd like to use your own API key, the best option is to clone the Ellide GitHub repository and run it locally. There are setup instructions in the README to get you up and running with your own credentials.

Bring-your-own-key support in the web app may be a future feature, but for now, self-hosting is the way to go.

That's a fair concern. The core engine behind Ellide is 100% open source under the MIT license.

If you have security requirements that go beyond what our hosted service offers, you can clone the repository, audit the code yourself, and run it on the cloud provider or local machine you trust most.

Yes — Ellide is free to use. You can process documents without creating an account, subject to a 50 MB file size limit.

Creating a free account raises the limit to 200 MB and lets you optionally save your cleaned outputs. Premium features like higher-tier AI models may be introduced later, but the core service will remain free.