Compare model answers, repeat a prompt under a named condition, keep the full run record, and ask a second model to inspect the first response.
Built for curious consumers and independent researchers who want a small, inspectable tool without bundled model weights or a cloud account.
Select up to three models, submit the same prompt, and inspect the answers together. Named experiments repeat the same input without prior conversation turns so each trial begins from a clean prompt.
After a primary answer completes, an optional verifier checks its claims and an optional reviewer critiques omissions, contradictions, unsupported leaps, and overconfidence.
Every event retains the prompt, response, provider, endpoint, model, experiment name, replicate number, status, timestamps, requested context, and provider-returned fields.
Unavailable values remain blank. Context size is shown as the requested setting, not a fabricated utilization meter.
The package contains a Python/Tkinter application and documentation. It does not include Python, model runtimes, or model weights. Install the provider separately and use models already on your device.
Settings and captures remain in %LOCALAPPDATA%\NuAegis\BLITZ Lite. There are no accounts, analytics, telemetry, sync, automatic updates, MCP tools, shell access, or code execution.
A non-local endpoint requires confirmation before BLITZ Lite sends the prompt, prior conversation context, or supplied evidence to it. Provider privacy rules then apply.
START_BLITZ_LITE.bat.Released under the MIT License. Personal and commercial use are allowed under the included license terms. No payment link or separate paid license is configured for this release.