Chat with PDF
Ask a PDF questions in plain language. You pick the AI provider, you bring the key — imisspdf never sees the file.
Unlike our other tools, this one sends your document content to the AI provider you choose (OpenAI, Anthropic, Google, or your own Ollama server). Your key and your file content never touch our servers — the request goes directly from your browser to the provider you pick. Don't use for documents that can't go to any third party. See why this BYOK pattern matters for chat-with-PDF privacy →
Select a PDF to chat with
or drop one here — text-based PDFs work best (scans need OCR first)
Connect your AI provider
Pick a provider and paste your key. Stored only in your browser's localStorage.
How it works
Pick a PDF
Drop or select one PDF. We extract its text locally in your browser.
Bring a key
Paste your API key from OpenAI, Anthropic, Google AI Studio, or use your local Ollama.
Ask away
Each question goes directly browser → provider → back. imisspdf never sees the file or the key.
Related PDF tools
What "Chat with PDF" means here
Chat with PDF turns a static document into something you can interrogate. You upload a PDF, the page extracts the text locally, and then every question you type is sent — along with the document text — to a large language model. The model reads the document and your question together and writes back an answer. Done well, this lets you ask things like "what does section 4 say about termination?" or "summarise the three risks listed on page 7" and get a useful response in a few seconds.
Our take is more honest than most: we tell you upfront that a third-party AI sees your document. We can\'t do otherwise — modern LLMs run on someone\'s GPU, not in your browser tab. What we CAN do is keep imisspdf out of the loop entirely. You pick the provider you trust, you pay for the usage, your key stays in your browser, and the request goes directly to the provider. We do not see the question, the answer, or the document.
Why "BYOK" (Bring Your Own Key)
A normal SaaS chat-with-PDF service runs the LLM call through their own server with their own key, and charges you a subscription. That model has two problems for privacy-minded users: their server sees your document on the way to the AI, and you have to trust their retention claims. We side-step both by inverting the architecture — you bring the key and we don\'t need a server step at all.
It does mean a small setup cost: you need to create an API key once, paste it in, and pick a model. After that, every question is one network call from your browser to the provider, billed to your account. For light use, this is far cheaper than a subscription (a few cents per dozen questions on cheap models like OpenAI\'s gpt-4o-mini). For heavy use, you can drop in Ollama and pay nothing at all.
Which provider should I pick?
- OpenAI (gpt-4o-mini, default): good baseline. Reliable, fast, ~$0.15 per 1M input tokens. Best when you just want it to work.
- Anthropic (Claude Haiku): very strong at long documents and careful reasoning. Roughly comparable price. Note: Anthropic\'s API doesn\'t officially support browser access — this tool sets an explicit opt-in header to allow it.
- Google (Gemini Flash): generous free tier for low-volume personal use. Fast.
- Ollama (self-hosted): zero cloud, zero cost beyond electricity. Install Ollama, run
ollama pull llama3.2, set the base URL tohttp://localhost:11434. The document and questions never leave your machine.
Privacy & security
Concretely: when you click Send, your browser sends a request to api.openai.com (or api.anthropic.com, generativelanguage.googleapis.com, or localhost:11434) containing your API key and the message stack — system prompt with the document text, your full chat history, your new question. The provider responds with the assistant\'s reply. That round-trip happens entirely between your browser and the provider. imisspdf serves you the HTML/JS once and is uninvolved afterwards.
Your API key is stored under localStorage["imisspdf.llm.key.<provider>"]. It\'s readable by any JS running on the imisspdf origin — which is to say, the JS we ship to make this page work. We do not exfiltrate keys; you can confirm by inspecting the Network tab. To wipe everything, click "Forget all keys" or clear site data in your browser.
Frequently asked questions
No. This tool sends the PDF text and your API key DIRECTLY from your browser to whichever AI provider you pick (OpenAI, Anthropic, Google, or your own Ollama server). imisspdf is not in the request path — there is no proxy, no relay, no edge function. Your key is stored in localStorage only on your device, and you can wipe it at any time with the "Forget all keys" button.
Every other tool on imisspdf runs entirely in your browser with no external dependency. Chat-with-PDF needs a large language model, and large language models live on servers (OpenAI's, Anthropic's, Google's) or on your own hardware (Ollama). We don't run an LLM ourselves and we don't want to insert ourselves between you and the provider — so you bring your own key and pay your own usage.
For most users: OpenAI gpt-4o-mini is cheap, fast, and quality is high. Anthropic's Claude is excellent at long documents and careful reasoning. Google Gemini Flash is free at low volumes (within Google's daily limits). Ollama is best if you want zero-cloud — install Ollama locally, pull a model like llama3.2, and point this tool at http://localhost:11434.
Depends on the document and provider. A typical 10-page PDF + a few questions runs $0.001-0.01 on OpenAI gpt-4o-mini. Big documents (hundreds of pages) cost more because the whole text is included in every question. With Ollama running locally, the cost is your electricity.
Only if you trust the provider. Even though imisspdf never sees the file, the AI provider does — that's how it answers your question. For documents that can't go to a third party at all, use Ollama on your own machine: the chat then never leaves your computer.