Precursor lets legal professionals safely use state of the art ai on medical records.

Why can’t I just use ChatGPT directly?

Uploading unredacted medical records into any cloud ai means transmitting confidential information to a third party. That’s a confidentiality concern most legal professionals are already aware of. Precursor removes the identifying information before anything leaves your machine, so you can use the same tools without the same exposure.

Where does the ai actually run?

Wherever you choose. Precursor’s job is to produce a safe input file. You take that file to Claude, ChatGPT, Gemini, a local model — whatever fits your workflow. Precursor doesn’t lock you in.

How do I use the redacted output to write summaries, draft expert reports, or answer complex medical questions?

Open the ai of your choice in your browser — ChatGPT (currently the strongest for medical-records analysis), Claude (close second), or Gemini if you don’t have access to either. Upload the files Precursor produced and input your request. The files Precursor produced are structured for the ai to navigate on its own: a set of chunked txt files plus a short index. Make your request in plain English: Write a chronological treatment summary, Draft the medical-history section of an expert report, or any of the example questions shown at the top. The workflow is the same across ai models: upload, ask, refine.

Why use Precursor instead of a legal tool that uploads medical records to its own ai?

Two reasons: model choice and confidentiality control.

All-in-one tools tie you to whichever model the vendor picked — which may not be the model best suited to your records, and may not stay the best for long. Precursor produces a clean redacted file and leaves the model choice to you. Claude, ChatGPT, Gemini, a local model — whatever fits the work today, and as better models emerge.

The other reason is confidentiality. Uploading unredacted records to any cloud ai — including a vendor’s own — hands that decision over to the vendor’s policies. Precursor keeps it with you: identifying information is stripped on your machine before anything is uploaded.

Ultimately, the confidentiality of records you put into a cloud ai is your responsibility — and that responsibility doesn’t transfer to any vendor. Precursor is built to help you meet it.

What gets sent to the cloud?

Only Precursor’s redacted, ai-ready file, and only when you choose to upload it.

How can I be confident the redacted file is safe to upload?

Precursor’s redaction is built on a de-identification system developed by the National Library of Medicine, well-established and widely used in healthcare research. It removes structured PHI patterns (names, dates, addresses, MRNs, account numbers, and more) and removes handwritten content. No automated redaction is perfect, which is why Precursor puts a human in the loop: you see the redacted output and can verify it before deciding to upload anywhere. The combination of a proven de-identification engine plus your final review is what makes the file safe to share with a cloud ai. A final-check function — built to independently verify the de-identification output — is now in development.

What records does it work on?

Today: medical records — provider charts, hospital records, billing files, itemized medical liens, explanation of benefits, etc.…

Coming soon: other categories of confidential records (e.g. financial records).

What happens if my records contain handwriting?

Precursor detects handwritten content and removes it from the output. Precursor can’t read handwriting and most ai models can’t reliably read handwriting. For records that are mostly typed — the vast majority of modern medical records — this works well. For records that are mostly handwritten, the output may be too sparse to be useful. If your case turns on handwritten content, Precursor isn’t the right tool for that content.

What other limits should I know about?

File formats: PDF only — both scanned and digital.

Language: English only.

Performance: No hard file-size or page-count limit — 1,000+ page PDFs are supported. Because Precursor runs on your computer, processing time scales with file size and scan quality. For very large files (above ~200 pages), some quality passes (like handwriting detection) are skipped to keep processing tractable. The practical ceiling is your machine’s compute power and disk space. Precursor runs on low-performance PCs commonly found in business settings. However, performance is limited to what those machines can handle.

Does it run on Mac?

Not yet. Precursor is currently a Windows desktop application. Mac and Linux are on the roadmap; sign up to be notified.