copilot-stats

A command-line tool that shows your GitHub Copilot usage statistics in USD, with full detail.

It combines two data sources that normally live far apart:

| Source | What it gives you | USD basis | | --- | --- | --- | | Local Copilot CLI session store (~/.copilot/session-store.db) | Every request you made: model, input/output/cache/reasoning tokens, AIU, latency, session, repo, branch | Estimated from configurable per-model rates | | GitHub billing & Copilot REST APIs | Org seat counts, premium-request line items, Copilot metrics | Real amounts billed by GitHub |

Important: local figures are estimates. GitHub Copilot is seat-licensed, so token-derived cost represents equivalent API spend (an opportunity cost), not an invoice. Numbers from the GitHub billing API are the real billed amounts.

Install

Requires Node.js 22.13+ (or 23.4+), which is where the built-in node:sqlite module became available without a flag. Zero runtime dependencies. Works on macOS, Linux and Windows.

git clone https://github.com/khanaffan/copilot-stats.git && cd copilot-stats
npm install
npm run build
npm link          # makes `copilot-stats` available globally

Or run it directly without linking:

node dist/cli.js

Quick start

copilot-stats                    # full summary of all local usage
copilot-stats models --since 30d # model costs for the last 30 days
copilot-stats sessions --top 10  # your ten most expensive sessions
copilot-stats org --org my-org   # org seat + premium-request spend

Commands

| Command | Description | | --- | --- | | local | (default) Personal usage from the local session store, costed in USD | | org | Org-level Copilot billing: seats and premium requests, real USD from GitHub | | all | Local and org reports together, with a combined total | | models | Per-model breakdown: tokens, requests, AIU, USD, share %, avg $/request | | sessions | Per-session breakdown: repo, branch, tokens, USD, duration, last used | | repos | Cost rolled up by repository or working directory | | daily | Time series by day, with --weekly / --monthly bucketing | | pricing | Show the effective pricing table; pricing init writes an override file | | help, version | Self-explanatory |

Options

-s, --since <date>      Start of window: YYYY-MM-DD, 7d, 4w, 6m, 1y, today, yesterday
-u, --until <date>      End of window (inclusive)
-m, --model <name>      Filter by model; repeatable or comma-separated
-r, --repo <name>       Filter by repository or working-directory substring
-o, --org <name>        Org for billing commands; defaults to all your orgs
    --currency <code>   Display currency (default USD); any code with an fxRates entry
-f, --format <fmt>      table | json | csv
    --json / --csv      Shorthands for --format
-n, --top <n>           Limit ranked table rows (or "all")
    --daily/--weekly/--monthly   Bucketing for the daily report
    --db <path>         Override the session store path
    --pricing <path>    Override the pricing config path
    --token <token>     GitHub token (else GITHUB_TOKEN, GH_TOKEN, gh auth token)
    --redact            Hash repository and path names in the output
    --no-color          Disable ANSI color
-v, --verbose           Show diagnostics, API calls and warnings

Example output

Sample data below is illustrative — repository, session and branch names are fabricated.

copilot-stats

Copilot Usage Summary

  $15,616.47  estimated cost
  13,334 requests across 13 models, 151 sessions
  Window: the beginning → now
  Billed request units: 118,932 (premium-multiplier weighted)
  Tokens: 1.3B in / 9.3M out / 1.2B cache-read / 49.7M cache-write / 3.6M reasoning (2.6B total)
  AIU: 99.5K
  Cache hit ratio: 48.7%  ·  Cache savings: $11,283.02
  Avg cost/request: $1.17
  Total duration: 37h 9m

Estimated from configurable rates — not a GitHub invoice.

Model                    Requests  Billed Reqs   Input  Output  Cache R  Cache W  Reasoning    AIU        Cost  Share %  Avg $/req
-----------------------  --------  -----------  ------  ------  -------  -------  ---------  -----  ----------  -------  ---------
claude-opus-4.8             5,264       78,610  528.3M    4.5M   502.0M    21.9M       2.0M  52.9K   $9,429.04    60.4%      $1.79
claude-opus-5                2,414       36,210  249.0M    1.7M   237.9M     9.2M     427.3K  23.1K   $4,391.99    28.1%      $1.82
claude-sonnet-5             4,213        4,213  412.5M    2.0M   393.5M    15.8M     652.9K  14.6K   $1,445.19     9.3%      $0.34
...
-----------------------  --------  -----------  ------  ------  -------  -------  ---------  -----  ----------  -------  ---------
TOTAL                      13,334      118,932    1.3B    9.3M     1.2B    49.7M       3.6M  99.5K  $15,616.47   100.0%      $1.17

The default report also includes tables by repository, by session, a daily trend with inline bar charts, and a breakdown by initiator (user vs. agent).

copilot-stats sessions --top 5

Session                            Branch  Requests  Tokens       Cost  Duration  Last used
---------------------------------  ------  --------  ------  ---------  --------  ----------
acme/checkout-service              main         464   99.6M    $826.53    1h 19m  2026-07-17
acme/billing-api                   main         426   95.2M    $824.83   54m 12s  2026-08-12
invoice-pdf-spike                  -            178   69.4M    $601.73   48m 37s  2026-07-08
---------------------------------  ------  --------  ------  ---------  --------  ----------
TOTAL (top 3 of 151)                          1,068  264.2M  $2,253.09     3h 2m

copilot-stats daily --weekly

Date        Requests  Tokens       Cost  Chart
----------  --------  ------  ---------  --------------------
2026-07-06     2,116  457.5M  $3,704.60  ████████████████████
2026-07-13     3,218  573.8M  $3,154.55  █████████████████░░░
2026-07-20     2,973  527.0M  $3,194.64  █████████████████░░░
2026-08-10     2,261  416.9M  $3,158.41  █████████████████░░░

Machine-readable output

copilot-stats models --json | jq '.data.rows[0].cost.total'
copilot-stats repos --csv > repo-costs.csv

JSON is wrapped in a stable envelope:

{ "tool": "copilot-stats", "version": "0.1.0", "kind": "models",
  "generatedAt": "2026-08-13T14:55:20.910Z", "data": { "...": "..." } }

Pricing configuration

Costs are computed from a bundled pricing file with sensible list-price defaults for every model the Copilot CLI uses. Inspect the effective rates at any time:

copilot-stats pricing

Create your own override, which is deep-merged over the defaults:

copilot-stats pricing init      # writes ~/.copilot-stats/pricing.json

{
  "usdPerAiu": 0.05,            // fallback rate for models with no token rates
  "seatPricePerMonth": 39,      // used to estimate org seat cost
  "fxRates": { "EUR": 0.92 },   // display currencies for --currency
  "models": {
    "claude-opus-5": {
      "input": 15,              // USD per 1,000,000 tokens
      "output": 75,
      "cacheRead": 1.5,
      "cacheWrite": 18.75
    }
  }
}

How a request is priced:

1. If the model has token rates configured for both input and output, cost = Σ (tokens ÷ 1,000,000 × rate) for input, output, cache-read and cache-write. Cache-read falls back to the input rate when unset; reasoning tokens are only billed separately if a distinct reasoning rate is configured (they are normally already counted inside output tokens). A model is only considered "token-priced" once both input and output are set — a partial override (e.g. just { "input": 3 }) would otherwise silently bill the rest of that model's usage (usually the bulk of the cost) at $0. Partially-configured models fall back to AIU pricing instead and are flagged in a distinct "partially priced" warning naming the missing keys.
2. Otherwise cost = AIU × usdPerAiu, and the row is marked ~ in the table with a footnote. A multi-model group (session/repo/day/total) that combines a token-priced and an AIU-priced model reports cost.basis: "mixed" (also shown with the ~ marker), rather than misleadingly collapsing to "tokens".

Each request also carries a request_multiplier from GitHub (>1 for premium requests). This is surfaced separately as billed request units (billedUnits in JSON/CSV, "Billed Reqs" in the table) — the sum of COALESCE(request_multiplier, 1) across the group. It is purely informational: it is not folded into token cost, since token cost is already computed directly from token counts and per-token rates.

Use --pricing <path> to point at a different file for a single run.

Currencies

All pricing is defined and computed in USD, which is also the default display currency. Pass --currency <code> to restate every monetary figure using the fxRates in the pricing config:

copilot-stats --currency EUR            # € formatted table
copilot-stats models --currency JPY --json
copilot-stats repos --currency GBP --csv > repos.csv

• Bundled rates: USD (always 1), EUR, GBP, CAD, AUD, INR, JPY.
• Add or override any code — including your own rate for an existing one — with an fxRates entry in ~/.copilot-stats/pricing.json. Codes are case-insensitive.
• copilot-stats pricing lists the currencies currently available, and an unknown code fails with an error naming them.
• Conversion is applied consistently to table, JSON and CSV output. JSON reports include a currency: { code, rate } field and CSV rows a cost_currency column, so exported numbers are never ambiguous.
• Rates are static config, not live FX — keep them current yourself if that matters.

Org billing and token scopes

Org commands need a token with elevated permissions. A standard gh auth login token does not have them, and GitHub returns 404 (not 403) when they are missing:

gh auth refresh -h github.com -s admin:org -s manage_billing:copilot

Without those scopes, copilot-stats org still runs and prints an explanatory warning per endpoint rather than failing, and copilot-stats all still shows your full local report. Endpoints used:

• GET /orgs/{org}/copilot/billing — plan type and seat breakdown
• GET /orgs/{org}/copilot/billing/seats — per-seat detail
• GET /organizations/{org}/settings/billing/usage — premium-request USD line items
• GET /orgs/{org}/copilot/metrics — active users, acceptance rates

Privacy

Nothing is ever sent anywhere except to GitHub's own API. The local session database is opened read-only (with WAL-safe fallbacks) and is never written to, so it is safe to run while a Copilot CLI session is active. Session summaries, repository names and working directory paths can be sensitive — use --redact to replace them with stable hashes when sharing output.

Development

npm run build      # compile to dist/
npm test           # compile and run the node:test suite (62 tests)
npm run dev        # tsc --watch

License

MIT