ContextQMD
Back to Promptfoo

Using the web viewer

site/docs/usage/web-ui.md

0.121.94.5 KB
Original Source

Using the web viewer

After running an eval, view results in your browser:

sh
npx promptfoo@latest view

See promptfoo view for CLI options.

Keyboard Shortcuts

ShortcutAction
Ctrl+K / Cmd+KOpen eval selector
EscClear search
Shift (hold)Show extra cell actions

Toolbar

  • Eval selector - Switch between evals
  • Zoom - Scale columns (50%-200%)
  • Display mode - Filter: All, Failures, Passes, Errors, Different, Highlights
  • Search - Text or regex
  • Filters - By metrics, metadata, pass/fail. Operators: =, contains, >, <

Table Settings

  • Columns - Toggle variable and prompt visibility
  • Truncation - Max text length, word wrap
  • Rendering - Markdown, JSON prettification
  • Inference details - Tokens, latency, cost, tokens/sec
  • Media - Image size limits; double-click for lightbox

Cell Actions

Hover to reveal actions. Hold Shift for more:

ActionDescription
🔍DetailsFull output, prompt, variables, grading results
👍PassMark as passed (score = 1.0)
👎FailMark as failed (score = 0.0)
🔢ScoreSet custom score (0-1)
✏️CommentAdd notes
⭐HighlightMark for review (Shift)
📋CopyCopy to clipboard (Shift)
🔗ShareLink to this output (Shift)

Ratings and comments persist and are included in exports—use them to build training datasets.

Eval Actions

  • Edit name - Rename eval
  • Edit and re-run - Open in eval creator
  • Compare - Diff against another eval (green = added, red = removed)
  • View YAML - Show config
  • Download - Opens export dialog:
ExportUse case
YAML configRe-run the eval
Failed tests onlyDebug failures
CSV / JSONAnalysis, reporting
DPO JSONPreference training data
Human Eval YAMLHuman labeling workflows
Burp payloadsSecurity testing (red team only)
  • Copy - Duplicate eval
  • Share - Generate URL (see Sharing)
  • Delete

Results Charts

Toggle with Show Charts.

Pass Rate

Percentage of tests where all assertions passed.

Score Distribution

Histogram of scores per prompt. Each test score = mean of its assertion scores. See weighted assertions.

Scatter Plot

Compare two prompts head-to-head. Click to select prompts.

  • Green = Prompt 2 scored higher
  • Red = Prompt 1 scored higher
  • Gray = Same score

Sharing

Eval actions → Share generates a URL.

Cloud

Free at promptfoo.app. Links are private to your organization.

sh
promptfoo auth login -k YOUR_API_KEY
promptfoo share

Self-hosted

For self-hosted deployments:

yaml
sharing:
  apiBaseUrl: http://your-server:3000
  appBaseUrl: http://your-server:3000

Or set via API Settings in the top-right menu. See sharing docs for auth and CI/CD.

URL Parameters

Viewer state syncs to the URL—bookmark or share filtered views:

ParameterValues
filterModeall, failures, passes, errors, different, highlights
searchAny text
text
/eval/abc123?filterMode=failures&search=timeout