Copilot CLI Cost adds estimated session-cost reporting to GitHub Copilot CLI.

It estimates usage-based billing with GitHub AI Credits.

The calculator stores canonical cost in USD and converts to a selected display currency with cached exchange rates from Frankfurter or an explicit exchange-rate override.

• /cost command for active-session estimates
• /cost session <session-id> for completed local sessions
• Native cost panel with token bucket breakdowns
• What-if subscription comparison for Copilot Free, Pro, Pro+, Max, Business, Enterprise, and Student
• Display currency selector backed by cached Frankfurter USD exchange rates
• Statusline cost segment with optional passthrough to another statusline
• Standalone calculator CLI for sample data, JSON files, completed session events, and live snapshots
• USD-first cost model with optional display currency conversion

Install by executing the install script for your platform.

Windows PowerShell (install.ps1):

irm https://raw.githubusercontent.com/DamianEdwards/copilot-cli-cost/main/install.ps1 | iex

macOS/Linux (install.sh):

bash -c "$(curl -fsSL https://raw.githubusercontent.com/DamianEdwards/copilot-cli-cost/main/install.sh)"

The remote scripts run in isolation and fetch their helper from the same raw-content base URL before configuring Copilot. To run the installer script from a checkout instead:

.\install.ps1

./install.sh

The installer:

• On Copilot CLI 1.0.56-0 or newer, registers this repository's marketplace and installs or updates copilot-cli-cost@copilot-cli-cost-marketplace. Older Copilot CLI versions use the direct repository install format.
• Downloads the installer helper from the raw-content base URL.
• Installs the user-scoped extension shim for /cost and the panel.
• Enables the Copilot experimental flags needed for extensions and the status line.
• Configures a stable user-scoped statusline launcher under ~/.copilot/copilot-cli-cost/.

If you already have a Copilot status line configured, the installer prompts you to replace it, decorate it with the Copilot Cost status line using passthrough mode, or skip statusline configuration. Before changing existing settings that would otherwise be overwritten, it prompts and writes a timestamped settings.json.bak-* backup.

Installer options:

Set COPILOT_COST_PLUGIN_SOURCE or pass --plugin-source <source> to install from a fork or alternate plugin source. On Copilot CLI 1.0.56-0 or newer, this source is registered as a plugin marketplace; set COPILOT_COST_MARKETPLACE_NAME / --marketplace-name and COPILOT_COST_PLUGIN_NAME / --plugin-name if your fork changes those marketplace identifiers. Set COPILOT_COST_INSTALL_BASE_URL or pass --install-base-url <url> when running installer scripts from an alternate raw-content location. Set COPILOT_HOME or pass --copilot-home <path> to isolate installer writes.

If /cost is not available in an active Copilot CLI session after installing, run /extensions and enable copilot-cli-cost under User.

When you run copilot inside this repository, Copilot CLI's extension resolver prefers the repo-local .github/extensions/copilot-cli-cost extension over the user-installed extension. The user-scoped shim intentionally stays pointed at the installed plugin copy for sessions outside the repository.

The scripts perform these steps. To do them manually, first install the plugin:

copilot plugin marketplace add DamianEdwards/copilot-cli-cost # if not already registered
copilot plugin install copilot-cli-cost@copilot-cli-cost-marketplace

Then run the extension shim installer from the installed plugin.

PowerShell terminal:

$installer = Get-ChildItem "$env:USERPROFILE\.copilot\installed-plugins" -Directory -Recurse |
  Where-Object { Test-Path (Join-Path $_.FullName "scripts\install-extension-shim.mjs") } |
  Select-Object -First 1 -ExpandProperty FullName

if (-not $installer) {
  throw "Could not find installed copilot-cli-cost plugin."
}

node (Join-Path $installer "scripts\install-extension-shim.mjs")

Bash terminal:

installer="$(find "$HOME/.copilot/installed-plugins" -type f -path '*/scripts/install-extension-shim.mjs' | head -n 1)"
if [ -z "$installer" ]; then
  echo "Could not find installed copilot-cli-cost plugin." >&2
  exit 1
fi
node "$installer"

Configure ~/.copilot/settings.json. Do not put statusLine in config.json; that file is managed by Copilot CLI and user settings may be moved or removed during startup. Use your machine's statusline launcher path:

{
  "experimental": true,
  "experimental_flags": ["EXTENSIONS", "STATUS_LINE"],
  "statusLine": {
    "type": "command",
    "command": "C:\\Users\\alex\\.copilot\\copilot-cli-cost\\statusline.cmd"
  },
  "footer": {
    "showCustom": true
  }
}

macOS/Linux:

{
  "experimental": true,
  "experimental_flags": ["EXTENSIONS", "STATUS_LINE"],
  "statusLine": {
    "type": "command",
    "command": "sh \"/Users/alex/.copilot/copilot-cli-cost/statusline.sh\""
  },
  "footer": {
    "showCustom": true
  }
}

The statusline bridge prints a compact segment:

💸 Cost ~$0.3059 (30.6 cr, 2% pro) · last 42K in/3K out

When the SDK extension can detect your current Copilot subscription, the statusline uses that cached plan for allowance percentages. COPILOT_COST_PLAN can still override the plan explicitly. If neither is available, the statusline falls back to assumed pro so the percentage is not presented as a detected plan.

The generated statusline launcher is workspace-aware because statusline settings point at a fixed command and do not use the extension resolver. When Copilot sends a statusline payload with workspace.current_dir or cwd inside a copilot-cli-cost checkout or worktree, the launcher runs that checkout's src/cli/statusline.js; otherwise it falls back to the installed plugin copy. Set COPILOT_COST_STATUSLINE_DISABLE_WORKSPACE=true to always use the installed copy.

/cost
/cost help
/cost version
/cost update
/cost panel on
/cost panel off
/cost panel refresh
/cost session <session-id>
/cost live-session <session-id>
/cost --plan pro|pro-plus|max|business|enterprise
/cost --currency USD|EUR|GBP|CAD|AUD|JPY|CHF

/cost is handled by extension JavaScript. It does not ask the model to calculate the result.

Use /cost update to force the extension to re-detect your current Copilot subscription and rewrite the shared cache used by the statusline. If the host exposes a statusline refresh hook, the command also requests an immediate refresh; otherwise the updated plan appears on the next normal statusline refresh.

The panel opens a native window:

/cost panel on

The panel shows:

• Usage-based estimate
• Percentage of the selected plan's allowance used by the session
• Searchable session picker for current, cached live, and completed sessions
• Selected session ID and data source
• Current or assumed subscription
• Loaded extension version
• What-if subscription selector
• Display currency selector
• Per-model token bucket breakdown
• Collapsed raw JSON payload

The SDK extension reads active-session metrics from Copilot CLI's session RPC API:

await session.rpc.usage.getMetrics()

That response includes:

• Per-model request counts
• Copilot-reported AI credit usage (totalNanoAiu) when available
• Input, cached input, cache write, output, and reasoning token buckets
• Active model
• Last-call input/output token counts
• API duration
• Code-change counters

The extension normalizes each read and writes a live snapshot to the platform cache folder:

Windows: %LOCALAPPDATA%\copilot-cli-cost\live-sessions
macOS:   ~/Library/Caches/copilot-cli-cost/live-sessions
Linux:   ${XDG_CACHE_HOME:-~/.cache}/copilot-cli-cost/live-sessions

Completed local sessions can be read from:

Windows: %USERPROFILE%\.copilot\session-state\<session-id>\events.jsonl
macOS/Linux: ~/.copilot/session-state/<session-id>/events.jsonl

The parser reads the latest metrics event and extracts Copilot-reported AI credit usage and per-model token buckets.

For usage-based billing, Copilot-reported AI credits are preferred because they match the CLI's own AI Credits counter. Token-rate estimates are retained as a fallback when Copilot does not provide AI credit totals, and the panel labels which method was used.

When statusline payloads include transcript_path, live snapshots are also grouped into a logical session. This keeps each resumed Copilot CLI instance as its own snapshot while letting /cost, the statusline segment, and the panel show the total cost across resumed instances.

Set COPILOT_COST_STATUSLINE_PASSTHROUGH to call another statusline command. The default passthrough mode enriches the stdin JSON with copilot_cost and lets the inner statusline render all output.

PowerShell:

$env:COPILOT_COST_STATUSLINE_PASSTHROUGH = "C:\Users\alex\.copilot\statusline\statusline.cmd"
copilot

macOS/Linux:

export COPILOT_COST_STATUSLINE_PASSTHROUGH="$HOME/.copilot/statusline/statusline.sh"
copilot

The enriched payload includes:

{
  "copilot_cost": {
    "schema_version": 2,
    "status_line": "💸 Cost ~$0.3059 (30.6 cr, 2% pro) · last 42K in/3K out",
    "aggregate_usage_based": {
      "billingModel": "usage-based",
      "totalUsd": 0.305869,
      "aiCredits": 30.5869,
      "creditCalculationSource": "copilot-cli-session-aiu"
    },
    "usage_based": {
      "billingModel": "usage-based",
      "totalUsd": 0.305869,
      "aiCredits": 30.5869,
      "creditCalculationSource": "copilot-cli-session-aiu"
    }
  }
}

Set decorate mode to combine this bridge's output with the passthrough output:

PowerShell:

$env:COPILOT_COST_STATUSLINE_MODE = "decorate"
$env:COPILOT_COST_STATUSLINE_POSITION = "right"
copilot

macOS/Linux:

export COPILOT_COST_STATUSLINE_MODE=decorate
export COPILOT_COST_STATUSLINE_POSITION=right
copilot

Statusline environment variables:

Set these environment variables before launching copilot:

PowerShell:

$env:COPILOT_COST_PLAN = "enterprise"
$env:COPILOT_COST_CURRENCY = "EUR"
copilot

macOS/Linux:

export COPILOT_COST_PLAN=enterprise
export COPILOT_COST_CURRENCY=EUR
copilot

The live session cache can be overridden with COPILOT_COST_LIVE_STORE. By default it uses the same platform cache root as COPILOT_COST_FX_CACHE.

The /cost command and calculator are pure Node.js. The native panel uses @webviewjs/webview, which installs a platform-specific optional package for Windows, macOS, and Linux x64. The extension bootstrap runs npm install --include=optional --no-audit --no-fund when those panel dependencies are missing.

On Linux, the native webview package still depends on system GTK/WebKit libraries supplied by your distribution, such as WebKitGTK and GTK. If /cost panel on fails to open, install your distribution's WebKitGTK/GTK runtime packages and reload the extension.

Clone the repository when you want to run tests or use the calculator directly:

git clone https://github.com/DamianEdwards/copilot-cli-cost.git
cd copilot-cli-cost
npm test

Examples:

npm run cost -- --sample
npm run cost -- --session <session-id> --plan pro
npm run cost -- --live --plan max
npm run cost -- --sample --currency EUR
npm run cost -- --sample --currency EUR --exchange-rate 0.9

Usage JSON shape:

{
  "sessionId": "sample-session-001",
  "plan": "pro",
  "currency": "USD",
  "modelUsage": [
    {
      "model": "gpt-5.5",
      "requests": 3,
      "inputTokens": 600000,
      "cachedInputTokens": 420000,
      "cacheWriteTokens": 0,
      "outputTokens": 36000,
      "reasoningTokens": 1200
    }
  ]
}

Usage-based billing uses published per-1M-token rates. inputTokens is the total input token count from Copilot metrics, and cachedInputTokens is the cached subset, so only uncached input tokens use the regular input rate:

uncachedInputTokens = max(inputTokens - cachedInputTokens, 0)
inputUsd            = uncachedInputTokens / 1,000,000 * inputPerMillionUsd
cachedInputUsd      = cachedInputTokens   / 1,000,000 * cachedInputPerMillionUsd
cacheWriteUsd       = cacheWriteTokens    / 1,000,000 * cacheWritePerMillionUsd
outputUsd           = outputTokens        / 1,000,000 * outputPerMillionUsd
reasoningUsd        = 0 unless COPILOT_COST_BILL_REASONING_TOKENS=true
aiCredits           = totalUsd / 0.01
includedAiCredits   = baseAiCredits + current flexAiCredits + current promotionalAiCredits

Some published model rates include a higher long-context tier. The calculator selects that tier when the model's total inputTokens exceeds the documented threshold, or when Copilot reports a model name that explicitly indicates an extended context variant such as Long Context, Large Context, or 1M Context.

For individual usage-based billing, Pro and Pro+ include a fixed base credit amount plus a variable flex allotment. The calculator reports the current published total as included credits and preserves the base/flex split in machine-readable output.

During GitHub's June 1-September 1, 2026 usage-based billing transition, existing Copilot Business and Enterprise customers receive promotional included credits. The calculator applies those promotional credits by default during that window and preserves them as a separate promotionalAiCredits component, for example 7,000 included (3,900 base + 3,100 promotional) for Copilot Enterprise.

Non-USD currency values are display estimates. USD remains canonical because GitHub model rates and AI Credits are documented in USD. Non-USD /cost and panel requests fetch USD exchange rates from Frankfurter and cache them for reuse; explicit environment or CLI exchange-rate overrides take precedence.

• Rate tables are hardcoded in src/core/rates.js and should be checked against GitHub billing docs.
• Reasoning tokens are shown as informational only unless COPILOT_COST_BILL_REASONING_TOKENS=true, because GitHub's published Copilot pricing table does not list a separate reasoning-token bucket.
• Business and Enterprise included credits are pooled at the billing entity level, so a session estimate is not always incremental billable spend.
• Taxes, regional billing rules, and GitHub billing-account currency handling are not modeled.
• Statusline per-model attribution depends on successive cumulative payloads and the active model at each refresh. Resumed-session aggregation depends on transcript_path; without it, live snapshots fall back to the current session_id and cannot be linked to previous instances.

MIT. See LICENSE.