GitHub's enterprise Copilot usage export is a per-user, per-day NDJSON stream of every
interaction, code suggestion, acceptance, language, feature, model, IDE and CLI token
burn. Drop your .ndjson file below and this page will parse it
entirely in your browser — nothing is uploaded — and turn it into a
summary you can scroll through, screenshot, or print.
Expected format: one JSON object per line (NDJSON / JSONL), as produced by the Copilot metrics export. Files up to a few hundred MB are fine — parsing is incremental.
.ndjson file hereHeadline figures across every row in the file.
user_login valuesday
Aggregated across every user, per day. Use this to spot weekday/weekend
patterns, holiday gaps, or rollout spikes.
Each row reports activity broken down by feature (chat panel, agent edit, agent mode, custom mode, CLI, etc). Bars sum the user-initiated interaction count for each feature across every row.
chat_panel_agent_mode dominates most enterprise exports.
Breakdown by model name as reported in totals_by_model_feature. This is the
single most useful view for spotting whether your org actually picks the model the
Pricing analysis predicts they should.
Code generation is bucketed by language. Lines added counts only the suggestions the user actually accepted into a file.
loc_added_sum. loc_suggested_to_add_sum shown for context.VS Code vs JetBrains vs Visual Studio vs the web — and where the CLI fits.
totals_by_ide.user_initiated_interaction_count per IDE name. CLI rows (which don't ship in totals_by_ide) are added as a synthesized "copilot-cli" surface.Top 15 users by total interactions. Row tooltips include their generations, acceptances, acceptance rate, and active days.
Only rows with used_cli: true contribute here. The CLI is the only surface
that publishes raw token counts in this export — useful for cross-referencing PRU
against actual provider token spend.
used_clisession_countused_cli: true, so the CLI charts above are empty.