TUI
BLXBench interactive Terminal User Interface.
BLXBench features a modern Terminal User Interface (TUI) built with Ink and React. It provides a command-driven experience for configuring providers, selecting models, running benchmarks, and submitting reports.
Install the CLI globally as @bitslix/blxbench (see Installation); the binary name remains blxbench.
Starting the TUI
blxbenchScreens
Welcome Screen
The welcome screen shown when starting BLXBench:
Help & Commands
View available commands and shortcuts:
Account Management
Sign in to upload results to the leaderboard:
Configure Run
Select provider, models, and test filters:
Running Benchmark
Watch real-time progress during execution:
Navigation
| Key | Action |
|---|---|
Tab | Complete slash-command suggestions |
Enter | Run the current command |
Ctrl+C | Exit |
After a benchmark finishes
The run dashboard stays open with a per-model summary and log. You can:
- Press
qorEscto go back to the shell (the benchmark does not re-run). - Press
sorrto upload the just-writtenreport.jsonto the public leaderboard, if you are signed in and your role is allowed to submit. This is a manual action: it works even when/report submitis off (auto-upload is separate).
The shell footer for an active run reminds you: s / r (public upload) and q / Esc (back).
Common Commands
| Command | Action |
|---|---|
/help | Show all commands by category |
/show | Show the active configuration |
/provider | Pick a provider |
/models list | Fetch models from the current provider |
/models id,id | Set model ids directly |
/categories * or /categories speed,reasoning | Select test categories |
/levels easy,medium,hard | Select difficulty levels |
/limit N | Limit tests per category |
/ratelimit RPM-or-off | Throttle provider requests |
| `/report html | json |
| `/report submit on | off` |
/resume | Pick a past report.json (same report root as the runner) to review or upload |
/output-dir PATH or /set output-dir PATH | Change the report directory |
/auth login | Sign in with browser device login |
/auth whoami | Show the signed-in account |
/playwright install | Install Playwright Chromium |
/run | Start the benchmark |
Account Login
/auth login starts the blxbench device flow. blxbench opens the web app, you approve the displayed code in the browser, and blxbench stores local credentials in your user config directory.
Leaderboard submission requires a signed-in account with a pass tier that includes submission quota. Headless automation can instead use BLXBENCH_API_KEY with --submit.
Local Reports
By default, TUI runs write reports under ~/.blxbench/reports/ on Linux/macOS and %USERPROFILE%\.blxbench\reports\ on Windows. Use /set output-dir PATH to override this for the current run. /report clear cleans the report directory while preserving/resetting ranking files.
/resume scans for recent report.json files under that effective path (default or your /set output-dir). If you change the output directory, resume uses the new location — it does not read from a different folder than the one your runs use.
After each public upload attempt (auto, s/r on a finished run, or from /resume), the CLI appends a local audit trail to the same report.json under the key blxbench_cli.public_submissions (timestamp, success or HTTP error, remote ids on success). That field is not sent to the server on upload, so you can see later what was published from this file.
Provider API keys remain local in your environment or .env. TUI sign-in is only for BLXBench account features such as eligible leaderboard upload.
Features
- Interactive model selection — Choose from available providers or type model ids directly
- Real-time progress — Watch benchmark execution
- Account integration — Sign in via browser device login and upload eligible reports
- Manual upload — After a run,
s/rupload the report without turning auto-submit on;/resumereopens older runs - Command help — Slash-command reference with completion