Introduction
Filling out PDF forms is a tedious chore that many professionals face daily, especially in industries like healthcare where document privacy is non-negotiable. While several AI-powered tools promise to streamline this process, they often compromise on security by sending sensitive data to remote servers. Enter SimplePDF Copilot—a revolutionary AI assistant that operates entirely within your browser, ensuring your confidential information never leaves your device. Built by the team behind SimplePDF, a privacy-respecting client-side PDF editor used by over 200,000 people monthly, Copilot brings intelligent form filling, field management, and page manipulation to your fingertips—all without sacrificing data sovereignty.
What Is SimplePDF Copilot?
SimplePDF Copilot is an AI-powered assistant embedded into the SimplePDF editor. Unlike conventional “Chat with PDF” tools that only retrieve text or OCR layers, Copilot can actively interact with the PDF document. It can:
- Fill fields automatically with context-aware answers
- Answer questions about the document’s content
- Focus on specific fields for user attention
- Add new fields detected client-side using CommonForms (by Joe Barrow) plus enhanced post-processing heuristics
- Delete pages and perform other document edits
The core idea is to provide a seamless, interactive experience where the AI behaves like an intelligent assistant that can take action on the PDF, not just read it.
Privacy-First Architecture
The most critical feature of SimplePDF Copilot is its privacy model. The PDF itself never leaves the browser. All parsing, rendering, and field detection happen client-side. Only the text the language model requires and your messages are sent to an LLM endpoint. By default, this is a demo proxy running DeepSeek V4 Flash (with rate limits), but users can bring your own key (BYOK) and point to any cloud provider. For ultimate privacy, you can even run the AI locally using tools like LM Studio.
This design was born from the needs of healthcare customers who handle protected health information (PHI). SimplePDF’s founder, who started the project seven years ago, wanted an AI experience that didn’t require shipping personally identifiable information (PII) to third parties.
The Technology: Client-Side Tool Calling
The magic behind Copilot lies in a technique called client-side tool calling. If you’re new to this concept, here’s a primer:
- Tool calling is how large language models (LLMs) take actions—like running commands, querying databases, or manipulating files. For example, when Claude uses
grepor connects to an MCP server, those are tool calls. - Client-side tool calling means the LLM expresses an intent to use a tool, but the actual execution happens in the user’s browser. This is achieved via an iframe and
postMessagecommunication.
This approach brings two major benefits:
- Speed: Client-to-client operations are the fastest possible—no round trips to remote servers for tool execution.
- Data control: You can limit what data the LLM sees. While the demo feeds document content to the AI, you can easily remove the tool that exposes that data, breaking the connection.
The implementation is open source and available on GitHub. However, the underlying SimplePDF editor (loaded as the iframe) remains proprietary.
Stack and Open Source
The tech stack is straightforward yet modern:
- Tanstack Start for the framework
- AI SDK from Vercel for LLM integration
- Tailwind CSS for styling (the founder admits he prefers CSS modules but chose Tailwind for broader community adoption)
The interesting part is the event-driven architecture: tool calls and responses flow between the main app and the SimplePDF iframe via postMessage, enabling real-time interactions.
Conclusion
SimplePDF Copilot demonstrates that AI-powered document handling doesn’t have to trade off privacy for convenience. By keeping the PDF local and executing tools in the browser, it achieves both speed and security. Whether you’re in healthcare, legal, or any field requiring strict data protection, Copilot offers a compelling alternative to cloud-dependent solutions. Try the live demo to see it in action, and explore the open-source code to understand the implementation details. For questions or contributions, the HN community thread (linked in the original post) is the place to be. The future of private, interactive PDF editing is here—and it runs entirely in your browser.