Version 1.1.0 · Built by Mo Shehu
A lightweight chargeback representment workbench. Takes a chargeback case, reads the relevant scheme reason-code rules, reviews attached merchant evidence, and produces an analyst-ready workup with a recommendation and evidence checklist.
Supports OpenAI and Anthropic models. Runs locally or behind a cloud deployment (Azure Container Apps or similar). No external services beyond the LLM API.
git clone https://github.com/shehuphd/casewright.git
cd casewrightCopy your evidence PDFs and images into data/documents/:
cp /path/to/evidence/files/* data/documents/Or set DOCUMENTS_DIR in .env to an absolute path elsewhere.
Copy .env.example to .env and fill in your key — or skip this and enter it via the in-app Settings panel on first launch.
cp .env.example .envSee USAGE.md for every .env field.
| Platform | How to run |
|---|---|
| macOS | Double-click launch.command in Finder |
| Linux | Run ./launch.sh in your terminal |
| Windows | Double-click launch.bat in Explorer |
On first run the launcher creates a virtual environment (outside the project folder, at ~/.venvs/casewright) and installs dependencies automatically. Subsequent launches skip straight to startup.
macOS note: if Gatekeeper blocks the script, right-click → Open the first time to approve it.
Windows note:
launch.batrequires Python to be on your PATH. Install from python.org and tick "Add to PATH" during setup.
The launcher finds a free port automatically (5050 by default, with fallback) and opens the app in your browser.
To reset all generated workups without losing case data:
python reset_outputs.py| Doc | Covers |
|---|---|
| USAGE.md | Configuration, deployment modes (local vs. bring-your-own-key cloud), document processing, the audit trail, LLM output shape, and known limitations |
| ARCHITECTURE.md | Request flow, module responsibilities, and how credential resolution differs between local and cloud mode |
| CHANGELOG.md | Version history |
Built by Mo Shehu