This guide walks you through everything needed to run ClaimJumper locally and see a complete denial triage result. By the end you will have cloned the repository, installed dependencies, configured your OpenAI API key, started the development server, uploaded the included sample EOB screenshot, and reviewed a ranked denial queue with cited draft artifacts.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/sputhenofficial/claimjumper/llms.txt
Use this file to discover all available pages before exploring further.
Prerequisites
Before you begin, make sure the following are available on your machine:
- Node.js
>=22 <23— the version range required by the project’senginesfield inpackage.json. - npm — bundled with Node.js; no separate install needed.
- OpenAI API key — ClaimJumper calls the OpenAI Responses API for vision ingest, denial triage, and draft generation. You can create a key at platform.openai.com/api-keys.
The repository includes a
.nvmrc file pinned to the supported Node.js version. If you use nvm, run nvm use in the project root to switch automatically.Clone and install
Clone the repository and install dependencies with npm.The install step downloads Next.js, React, the OpenAI SDK, Zod, Vitest, and the other dependencies declared in
package.json. There is no database to configure and no external service to provision.Configure your API key
Copy the example environment file to create your local configuration, then open Your ClaimJumper reads the key from
.env.local and replace the placeholder with your real key..env.local file should look like this:process.env.OPENAI_API_KEY on the server. It is never exposed to the browser.Start the development server
Start the Next.js development server with the following command.Next.js will print the local URL (typically
http://localhost:3000) to the terminal once the server is ready. Open that URL in your browser to see the ClaimJumper upload interface.Upload the sample EOB
The repository includes
samples/eob-001.png — a demo remittance screenshot that exercises all five denial routing lanes. Click the upload area on the home page, or drag and drop the file onto it, then click Triage.The app sends the image to POST /api/triage, where the server-side pipeline runs:- Ingest — GPT-5.6 Sol reads the PNG and extracts structured claim and service-line data.
- Triage — GPT-5.6 Sol routes each denied line to a lane and scores confidence and recovery probability. Code-enforced invariants run after the model decision.
- Prioritize — a deterministic function computes appeal deadlines and priority scores and sorts the queue.
- Draft — GPT-5.6 Terra produces cited appeal letters and corrected-claim worklists for eligible lines; human-review lines receive a deterministic safety brief.
appeal, corrected_claim, patient_bill, write_off, or human_review), appeal deadline, and priority score. Expand any row to read the rationale, evidence needed, routing confidence, and any override reasons. Appeal and corrected-claim rows include their draft artifact. Human-review rows show the safety brief with explicit override reasons.When you are ready to act on eligible drafts, click Approve on an appeal or corrected-claim row. Then click Download approved drafts to save a citation-preserving plain-text bundle for your billing team’s next controlled step.Verify the project
Before making any changes, confirm that the project passes type-checking, unit tests, and the production build.npm test runs the Vitest regression suite, which includes invariant coverage and pipeline tests using the sample fixtures.
For a guided, step-by-step walkthrough of a full triage session — including how to inspect each lane, read citation chips, approve drafts, and download the bundle — see the Demo Walkthrough.