@dianshuv/copilot-api 0.16.0 → 0.17.0

package/dist/main.mjs

@@ -1010,7 +1010,7 @@ const logout = defineCommand({
 
 //#endregion
 //#region package.json
-var version = "0.16.0";
+var version = "0.17.0";
 
 //#endregion
 //#region src/lib/event-loop-lag.ts
@@ -2715,6 +2715,20 @@ function formatTokens(input, output) {
 	return `${formatNumber(input)}/${formatNumber(output)}`;
 }
 /**
+* Cache-hit segment for the complete line, e.g. `cache=90%` (or
+* `cache=90% (1.6K/1.8K)` under --verbose). Returns undefined when there is no
+* usable denominator or the request had zero cache hits, so non-cached and
+* non-LLM requests stay terse (minimize-noise). `total` is the full prompt size
+* normalized across API families; see completeTracking / the per-path extraction.
+*/
+function formatCacheHitRate(cached, total) {
+	if (total === void 0 || total <= 0) return void 0;
+	if (cached === void 0 || cached <= 0) return void 0;
+	const pct = Math.round(cached / total * 100);
+	if (consola.level >= 5) return `cache=${pct}% (${formatNumber(cached)}/${formatNumber(total)})`;
+	return `cache=${pct}%`;
+}
+/**
 * Console renderer that shows request lifecycle with apt-get style footer
 *
 * Log format:
@@ -2816,7 +2830,7 @@ var ConsoleRenderer = class {
 	* Format a complete log line with colored parts
 	*/
 	formatLogLine(parts) {
-		const { prefix, time, method, path, model, status, duration, tokens, queueWait, phases, extra, isError, isDim } = parts;
+		const { prefix, time, method, path, model, status, duration, tokens, cache, queueWait, phases, extra, isError, isDim } = parts;
 		if (isDim) {
 			const modelPart = model ? ` ${model}` : "";
 			const extraPart = extra ? ` ${extra}` : "";
@@ -2830,6 +2844,7 @@ var ConsoleRenderer = class {
 		if (duration) result += ` ${pc.yellow(duration)}`;
 		if (queueWait) result += ` ${pc.dim(`(queued ${queueWait})`)}`;
 		if (tokens) result += ` ${pc.blue(tokens)}`;
+		if (cache) result += ` ${pc.dim(cache)}`;
 		if (phases) result += ` ${pc.dim(phases)}`;
 		if (extra) result += isError ? pc.red(extra) : extra;
 		return result;
@@ -2901,6 +2916,7 @@ var ConsoleRenderer = class {
 		const status = request.statusCode ?? 0;
 		const isError = request.status === "error" || status >= 400;
 		const tokens = request.model ? formatTokens(request.inputTokens, request.outputTokens) : void 0;
+		const cache = formatCacheHitRate(request.cachedInputTokens, request.totalInputTokens);
 		const queueWait = request.queueWaitMs && request.queueWaitMs > 100 ? formatDuration(request.queueWaitMs) : void 0;
 		const message = this.formatLogLine({
 			prefix: isError ? "[FAIL]" : "[ OK ]",
@@ -2912,6 +2928,7 @@ var ConsoleRenderer = class {
 			duration: formatDuration(request.durationMs ?? 0),
 			queueWait,
 			tokens,
+			cache,
 			phases: this.formatPhases(request),
 			extra: isError && request.error ? `: ${request.error}` : void 0,
 			isError,
@@ -2980,6 +2997,8 @@ var RequestTracker = class {
 		if (update.inputTokens !== void 0) request.inputTokens = update.inputTokens;
 		if (update.outputTokens !== void 0) request.outputTokens = update.outputTokens;
 		if (update.reasoningTokens !== void 0) request.reasoningTokens = update.reasoningTokens;
+		if (update.cachedInputTokens !== void 0) request.cachedInputTokens = update.cachedInputTokens;
+		if (update.totalInputTokens !== void 0) request.totalInputTokens = update.totalInputTokens;
 		if (update.error !== void 0) request.error = update.error;
 		if (update.queuePosition !== void 0) request.queuePosition = update.queuePosition;
 		if (update.queueWaitMs !== void 0) request.queueWaitMs = update.queueWaitMs;
@@ -4147,13 +4166,15 @@ function updateTrackerStatus(trackingId, status) {
 	requestTracker.updateRequest(trackingId, { status });
 }
 /** Complete TUI tracking and send PostHog analytics */
-function completeTracking(trackingId, inputTokens, outputTokens, queueWaitMs, reasoningTokens, analytics, timings) {
+function completeTracking(trackingId, inputTokens, outputTokens, queueWaitMs, reasoningTokens, analytics, timings, cache) {
 	if (!trackingId) return;
 	requestTracker.updateRequest(trackingId, {
 		inputTokens,
 		outputTokens,
 		queueWaitMs,
 		reasoningTokens,
+		cachedInputTokens: cache?.cachedInputTokens,
+		totalInputTokens: cache?.totalInputTokens,
 		...timingsToUpdate(timings)
 	});
 	requestTracker.completeRequest(trackingId, 200, {
@@ -4265,6 +4286,9 @@ function formatClientTruncationMarker(result) {
 function getReasoningTokensFromOpenAIUsage(usage) {
 	return usage?.completion_tokens_details?.reasoning_tokens;
 }
+function getCachedTokensFromOpenAIUsage(usage) {
+	return usage?.prompt_tokens_details?.cached_tokens;
+}
 async function handleCompletion$1(c) {
 	const rawPayload = await c.req.json();
 	consola.debug("Request payload:", JSON.stringify(rawPayload).slice(-400));
@@ -4383,7 +4407,9 @@ function handleNonStreamingResponse$1(c, originalResponse, ctx, payload) {
 		inputTokens: usage.prompt_tokens,
 		outputTokens: usage.completion_tokens,
 		queueWaitMs: ctx.queueWaitMs,
-		reasoningTokens
+		reasoningTokens,
+		cachedInputTokens: getCachedTokensFromOpenAIUsage(usage),
+		totalInputTokens: usage.prompt_tokens
 	});
 	captureRequest({
 		model: response.model,
@@ -4424,6 +4450,7 @@ function createStreamAccumulator() {
 		model: "",
 		inputTokens: 0,
 		outputTokens: 0,
+		cachedTokens: 0,
 		reasoningTokens: 0,
 		finishReason: "",
 		content: "",
@@ -4467,7 +4494,10 @@ async function handleStreamingResponse$1(opts) {
 			durationMs: Date.now() - ctx.startTime,
 			stopReason: acc.finishReason || void 0,
 			toolCount: payload.tools?.length ?? 0
-		}, ctx.timings);
+		}, ctx.timings, {
+			cachedInputTokens: acc.cachedTokens,
+			totalInputTokens: acc.inputTokens
+		});
 	} catch (error) {
 		if (isAbortError(error)) {
 			consola.debug("[ChatCompletions] client disconnected mid-stream; upstream aborted");
@@ -4536,6 +4566,7 @@ function accumulateParsedChunk(parsed, acc, checkRepetition) {
 	if (parsed.usage) {
 		acc.inputTokens = parsed.usage.prompt_tokens;
 		acc.outputTokens = parsed.usage.completion_tokens;
+		acc.cachedTokens = parsed.usage.prompt_tokens_details?.cached_tokens ?? 0;
 		acc.reasoningTokens = getReasoningTokensFromOpenAIUsage(parsed.usage) ?? 0;
 	}
 	const choice = parsed.choices?.[0];
@@ -7149,6 +7180,8 @@ function createAnthropicStreamAccumulator() {
 		model: "",
 		inputTokens: 0,
 		outputTokens: 0,
+		cacheReadInputTokens: 0,
+		cacheCreationInputTokens: 0,
 		stopReason: "",
 		content: "",
 		toolCalls: [],
@@ -7158,6 +7191,9 @@ function createAnthropicStreamAccumulator() {
 }
 function processAnthropicEvent(event, acc) {
 	switch (event.type) {
+		case "message_start":
+			handleMessageStart(event.message.usage, acc);
+			break;
 		case "content_block_delta":
 			handleContentBlockDelta(event.delta, acc);
 			break;
@@ -7193,11 +7229,20 @@ function handleContentBlockStop(acc) {
 		acc.currentToolCall = null;
 	}
 }
+function applyPromptSideUsage(usage, acc) {
+	if (usage.input_tokens !== void 0) acc.inputTokens = usage.input_tokens;
+	if (usage.cache_read_input_tokens !== void 0) acc.cacheReadInputTokens = usage.cache_read_input_tokens;
+	if (usage.cache_creation_input_tokens !== void 0) acc.cacheCreationInputTokens = usage.cache_creation_input_tokens;
+}
+function handleMessageStart(usage, acc) {
+	if (!usage) return;
+	applyPromptSideUsage(usage, acc);
+}
 function handleMessageDelta(delta, usage, acc) {
 	if (delta.stop_reason) acc.stopReason = delta.stop_reason;
 	if (usage) {
-		acc.inputTokens = usage.input_tokens ?? 0;
 		acc.outputTokens = usage.output_tokens;
+		applyPromptSideUsage(usage, acc);
 	}
 }
 function recordAnthropicStreamingResponse(acc, fallbackModel, ctx) {
@@ -8400,11 +8445,17 @@ function handleDirectAnthropicNonStreamingResponse(c, response, ctx, truncateRes
 		},
 		toolCalls: extractToolCallsFromContent(response.content)
 	}, Date.now() - ctx.startTime);
-	if (ctx.trackingId) requestTracker.updateRequest(ctx.trackingId, {
-		inputTokens: response.usage.input_tokens,
-		outputTokens: response.usage.output_tokens,
-		queueWaitMs: ctx.queueWaitMs
-	});
+	if (ctx.trackingId) {
+		const cacheRead = response.usage.cache_read_input_tokens ?? 0;
+		const cacheCreation = response.usage.cache_creation_input_tokens ?? 0;
+		requestTracker.updateRequest(ctx.trackingId, {
+			inputTokens: response.usage.input_tokens,
+			outputTokens: response.usage.output_tokens,
+			queueWaitMs: ctx.queueWaitMs,
+			cachedInputTokens: cacheRead,
+			totalInputTokens: response.usage.input_tokens + cacheRead + cacheCreation
+		});
+	}
 	captureRequest({
 		model: response.model,
 		inputTokens: response.usage.input_tokens,
@@ -8498,7 +8549,10 @@ async function handleDirectAnthropicStreamingResponse(opts) {
 			durationMs: Date.now() - ctx.startTime,
 			stopReason: acc.stopReason || void 0,
 			toolCount: anthropicPayload.tools?.length ?? 0
-		}, ctx.timings);
+		}, ctx.timings, {
+			cachedInputTokens: acc.cacheReadInputTokens,
+			totalInputTokens: acc.inputTokens + acc.cacheReadInputTokens + acc.cacheCreationInputTokens
+		});
 	} catch (error) {
 		if (isAbortError(error)) {
 			consola.debug("[Anthropic] client disconnected mid-stream; upstream aborted");
@@ -8713,11 +8767,17 @@ function handleNonStreamingResponse(opts) {
 		},
 		toolCalls: extractToolCallsFromContent(anthropicResponse.content)
 	}, Date.now() - ctx.startTime);
-	if (ctx.trackingId) requestTracker.updateRequest(ctx.trackingId, {
-		inputTokens: anthropicResponse.usage.input_tokens,
-		outputTokens: anthropicResponse.usage.output_tokens,
-		queueWaitMs: ctx.queueWaitMs
-	});
+	if (ctx.trackingId) {
+		const cacheRead = anthropicResponse.usage.cache_read_input_tokens ?? 0;
+		const cacheCreation = anthropicResponse.usage.cache_creation_input_tokens ?? 0;
+		requestTracker.updateRequest(ctx.trackingId, {
+			inputTokens: anthropicResponse.usage.input_tokens,
+			outputTokens: anthropicResponse.usage.output_tokens,
+			queueWaitMs: ctx.queueWaitMs,
+			cachedInputTokens: cacheRead,
+			totalInputTokens: anthropicResponse.usage.input_tokens + cacheRead + cacheCreation
+		});
+	}
 	captureRequest({
 		model: anthropicResponse.model,
 		inputTokens: anthropicResponse.usage.input_tokens,
@@ -8762,7 +8822,10 @@ async function handleStreamingResponse(opts) {
 			durationMs: Date.now() - ctx.startTime,
 			stopReason: acc.stopReason || void 0,
 			toolCount: anthropicPayload.tools?.length ?? 0
-		}, ctx.timings);
+		}, ctx.timings, {
+			cachedInputTokens: acc.cacheReadInputTokens,
+			totalInputTokens: acc.inputTokens + acc.cacheReadInputTokens + acc.cacheCreationInputTokens
+		});
 	} catch (error) {
 		if (isAbortError(error)) {
 			consola.debug("[Translated] client disconnected mid-stream; upstream aborted");
@@ -9395,7 +9458,10 @@ const handleResponses = async (c) => {
 							stream: true,
 							durationMs: Date.now() - startTime,
 							toolCount: tools.length
-						}, ctx.timings);
+						}, ctx.timings, {
+							cachedInputTokens: usage?.input_tokens_details?.cached_tokens ?? 0,
+							totalInputTokens: usage?.input_tokens ?? 0
+						});
 					} else if (streamErrorMessage) {
 						recordResponse(historyId, {
 							success: false,
@@ -9446,7 +9512,10 @@ const handleResponses = async (c) => {
 			stream: false,
 			durationMs: Date.now() - startTime,
 			toolCount: tools.length
-		}, ctx.timings);
+		}, ctx.timings, {
+			cachedInputTokens: usage?.input_tokens_details?.cached_tokens ?? 0,
+			totalInputTokens: usage?.input_tokens ?? 0
+		});
 		consola.debug("Forwarding native Responses result:", JSON.stringify(result).slice(-400));
 		return c.json(echoResponseBody(result, ctx));
 	} catch (error) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dianshuv/copilot-api",
-  "version": "0.16.0",
+  "version": "0.17.0",
   "description": "Turn GitHub Copilot into OpenAI/Anthropic API compatible server. Usable with Claude Code!",
   "author": "dianshuv",
   "type": "module",