Refresh pricing dynamically from GitHub Docs (#9)

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot-Session: 1adb4444-9149-4f9b-97b6-6dcd0847ba45

Author: devartifex

README.md

@@ -122,7 +122,7 @@ GitHub Copilot CLI ──OTel spans (JSONL)──▶ ~/.copilot/otel/copilot-ote
    export COPILOT_OTEL_FILE_EXPORTER_PATH="$HOME/.copilot/otel/copilot-otel.jsonl"
    ```
 2. **Aggregate** — on every render, recent spans are read, `gen_ai.usage.*` counters and `gen_ai.request.model` are extracted, and rolled up by session / model / day.
-3. **Estimate** — token counts are multiplied by a bundled pricing snapshot from GitHub's public [Copilot models & pricing](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing) table to estimate USD and GitHub AI Credits (AIC). This is not billing data; run `copilot-cost refresh-pricing --force` to refresh the local pricing cache when GitHub updates its published prices.
+3. **Estimate** — token counts are multiplied by prices from GitHub's public [Copilot models & pricing](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing) table to estimate USD and GitHub AI Credits (AIC). The local cache refreshes automatically when it is older than seven days, retains the last-known-good data if GitHub is unavailable, and falls back to the bundled snapshot when no cache exists. Long-context rates are selected from the published input-token thresholds. This is not billing data.
 4. **Render** — a one-line statusline, plus an optional local web dashboard.
 
 More on the underlying telemetry pipeline: [Copilot OpenTelemetry observability](https://docs.github.com/en/copilot/how-tos/copilot-sdk/observability/opentelemetry).
@@ -134,7 +134,7 @@ More on the underlying telemetry pipeline: [Copilot OpenTelemetry observability]
 - ✅ Usage data **stays on your machine**.
 - ✅ The dashboard only supports local binds (`127.0.0.1` or `localhost`).
 - ✅ This package emits **no telemetry or analytics** of its own.
-- 🌐 Pricing refresh contacts `docs.github.com` only when requested or when the cache needs refreshing.
+- 🌐 Pricing refresh contacts GitHub's public Docs pricing source only when requested or when the cache needs refreshing.
 
 ---
 
@@ -147,7 +147,7 @@ More on the underlying telemetry pipeline: [Copilot OpenTelemetry observability]
 | `uninstall [--yes]` | Remove settings installed by this package when they point at this tool. |
 | `doctor` | Check statusline setup, OpenTelemetry output, pricing, and dashboard readiness. |
 | `dashboard [--port <n>] [--host <h>] [--no-open]` | Serve the local dashboard. |
-| `refresh-pricing [--force]` | Refresh model pricing; `--force` bypasses the cache TTL. |
+| `refresh-pricing [--force]` | Refresh model pricing now; `--force` bypasses the seven-day cache TTL. |
 
 ---
 
@@ -157,7 +157,7 @@ More on the underlying telemetry pipeline: [Copilot OpenTelemetry observability]
 - **No usage shows up yet.** Make sure you're on the latest Copilot CLI, restart your shell and `copilot`, send a prompt, then check for JSONL files in `~/.copilot/otel/`.
 - **I do not want profile edits.** Use `copilot-cost install --no-otel-profile` and paste the printed OpenTelemetry block into the shell profile you choose.
 - **The dashboard will not bind.** Use a local host only, e.g. `copilot-cost dashboard --host 127.0.0.1 --port 4567`.
-- **Pricing looks stale.** Run `copilot-cost refresh-pricing --force`.
+- **Pricing looks stale.** Run `copilot-cost refresh-pricing --force`. Set `COPILOT_COST_REFRESH_DAYS` to change the automatic refresh interval, or `COPILOT_COST_AUTO_REFRESH=0` to disable automatic refreshes.
 
 ---
 

package-lock.json

@@ -61,6 +61,6 @@
     "tailwindcss": "^3.4.13",
     "tsup": "^8.3.0",
     "typescript": "^5.6.2",
-    "vitest": "^3.2.4"
+    "vitest": "^3.2.7"
   }
 }

package.json

@@ -13,9 +13,15 @@ async function readStdin(): Promise<string> {
   return Buffer.concat(chunks).toString("utf-8");
 }
 
+async function refreshStalePricing(): Promise<void> {
+  if (process.env.COPILOT_COST_PRICING || process.env.COPILOT_COST_AUTO_REFRESH === "0") return;
+  await refreshPricing();
+}
+
 async function renderCommand(): Promise<void> {
   try {
     const payload = JSON.parse(await readStdin()) as unknown;
+    await refreshStalePricing();
     console.log(renderPayload(payload));
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error);
@@ -54,6 +60,7 @@ export async function main(argv = process.argv.slice(2)): Promise<void> {
     .option("--host <host>", "host to listen on")
     .option("--no-open", "do not open the dashboard in a browser")
     .action(async (opts: { port?: number; host?: string; open?: boolean }) => {
+      await refreshStalePricing();
       await cmdDashboard({ port: opts.port, host: opts.host, noOpen: opts.open === false });
     });
   try {

src/cli.ts

@@ -111,7 +111,8 @@ export function normalizeSpan(record: unknown): NormalizedCall | null {
   if (!isObject(record) || !isChatSpan(record)) return null;
   const a = attrs(record);
   const rawModel = str(a["gen_ai.request.model"] ?? a["gen_ai.response.model"] ?? a["model"] ?? record.model);
-  const model = normalizeModel(rawModel) ?? rawModel;
+  const pricedModel = getModelPrice(rawModel);
+  const model = pricedModel.price ? pricedModel.model : normalizeModel(rawModel) ?? rawModel;
   if (!model) return null;
 
   const rawInput = num(a["gen_ai.usage.input_tokens"], a.input_tokens, record.input_tokens);
@@ -134,7 +135,7 @@ export function normalizeSpan(record: unknown): NormalizedCall | null {
   const responseId = str(a["gen_ai.response.id"] ?? record["gen_ai.response.id"]);
   const existingDedup = str(record.dedup_key);
   const dedupKey = existingDedup ?? (traceId && spanId ? `${traceId}:${spanId}` : responseId ?? (isLogRecord ? `${model}:${JSON.stringify(record.hrTime)}` : spanId ?? hashRecord(record)));
-  const { price } = getModelPrice(model);
+  const { price } = pricedModel;
   const usdCost = price ? computeCost({ input: freshInput + cacheRead + cacheCreation, cache_read: cacheRead, cache_write: cacheCreation, output }, price) : 0;
 
   return {