ghc-proxy 0.4.2 → 0.5.1

package/dist/main.mjs

@@ -5380,7 +5380,12 @@ const configFileSchema = object({
 	useFunctionApplyPatch: boolean().optional(),
 	responsesApiContextManagementModels: array(string()).optional(),
 	modelReasoningEfforts: record(string(), reasoningEffortSchema).optional(),
-	contextUpgrade: boolean().optional()
+	modelRewrites: array(object({
+		from: string(),
+		to: string()
+	})).optional(),
+	contextUpgrade: boolean().optional(),
+	contextUpgradeTokenThreshold: number().int().positive().optional()
 }).passthrough();
 const KNOWN_CONFIG_KEYS = new Set(Object.keys(configFileSchema.shape));
 let cachedConfig = {};
@@ -5388,6 +5393,7 @@ const DEFAULT_REASONING_EFFORT = "high";
 const DEFAULT_USE_FUNCTION_APPLY_PATCH = true;
 const DEFAULT_COMPACT_USE_SMALL_MODEL = false;
 const DEFAULT_CONTEXT_UPGRADE = true;
+const DEFAULT_CONTEXT_UPGRADE_TOKEN_THRESHOLD = 16e4;
 async function readConfig() {
 	try {
 		const content = await fs.readFile(PATHS.CONFIG_PATH, "utf8");
@@ -5439,6 +5445,9 @@ function isResponsesApiContextManagementModel(model) {
 function shouldContextUpgrade() {
 	return cachedConfig.contextUpgrade ?? DEFAULT_CONTEXT_UPGRADE;
 }
+function getContextUpgradeTokenThreshold() {
+	return cachedConfig.contextUpgradeTokenThreshold ?? DEFAULT_CONTEXT_UPGRADE_TOKEN_THRESHOLD;
+}
 function getReasoningEffortForModel(model) {
 	return cachedConfig.modelReasoningEfforts?.[model] ?? DEFAULT_REASONING_EFFORT;
 }
@@ -6216,7 +6225,7 @@ const checkUsage = defineCommand({
 
 //#endregion
 //#region src/lib/version.ts
-const VERSION = "0.4.2";
+const VERSION = "0.5.1";
 
 //#endregion
 //#region src/debug.ts
@@ -46561,12 +46570,20 @@ function colorizeMethod(method) {
 }
 function formatModelMapping(info) {
 	if (!info) return "";
-	const { originalModel, mappedModel } = info;
-	if (!originalModel && !mappedModel) return "";
-	const original = originalModel ?? "-";
-	const mapped = mappedModel ?? "-";
-	if (original === mapped) return ` ${colorize("dim", "model=")}${colorize("blueBright", original)}`;
-	return ` ${colorize("dim", "model=")}${colorize("blueBright", original)} ${colorize("dim", "→")} ${colorize("greenBright", mapped)}`;
+	const { originalModel, rewrittenModel, mappedModel } = info;
+	if (!originalModel && !rewrittenModel && !mappedModel) return "";
+	const parts = [];
+	const displayOriginal = originalModel ?? "-";
+	parts.push(colorize("blueBright", displayOriginal));
+	if (rewrittenModel && rewrittenModel !== displayOriginal) {
+		parts.push(colorize("dim", "~>"));
+		parts.push(colorize("cyanBright", rewrittenModel));
+	}
+	if (mappedModel && mappedModel !== (rewrittenModel ?? displayOriginal)) {
+		parts.push(colorize("dim", "→"));
+		parts.push(colorize("greenBright", mappedModel));
+	}
+	return ` ${colorize("dim", "model=")}${parts.join(" ")}`;
 }
 /**
 * Request logging function.
@@ -46585,6 +46602,19 @@ function logRequest(method, url, status, elapsed, modelInfo) {
 	console.log(`${line}${formatModelMapping(modelInfo)}`);
 }
 
+//#endregion
+//#region src/lib/request-timeout.ts
+function disableIdleTimeout(server, request) {
+	if (typeof server?.timeout === "function") server.timeout(request, 0);
+}
+function hasStreamingFlag(body) {
+	if (!body || typeof body !== "object") return false;
+	return body.stream === true;
+}
+function hasStreamingResponsesQuery(request) {
+	return new URL(request.url).searchParams.get("stream") === "true";
+}
+
 //#endregion
 //#region src/lib/sse-adapter.ts
 /**
@@ -47208,10 +47238,10 @@ var AnthropicStreamTranslator = class {
 	}
 	onChunk(chunk) {
 		const deltas = this.toConversationDeltas(chunk);
+		if (chunk.usage) this.state.lastUsage = chunk.usage;
 		if (deltas.length === 0) return [];
 		const events = [];
 		this.appendMessageStart(events, chunk);
-		this.state.lastUsage = chunk.usage;
 		for (const delta of deltas) switch (delta.kind) {
 			case "message_start": break;
 			case "thinking_delta":
@@ -47244,8 +47274,7 @@ var AnthropicStreamTranslator = class {
 					...delta.metadata
 				};
 				this.state.pendingStopReason = delta.stopReason;
-				if (delta.usage) this.state.lastUsage = delta.usage;
-				events.push(...this.onDone());
+				this.closeAllBlocks(events);
 				break;
 		}
 		return events;
@@ -47253,9 +47282,7 @@ var AnthropicStreamTranslator = class {
 	onDone() {
 		if (!this.state.messageStartSent || this.state.messageStopSent) return [];
 		const events = [];
-		this.thinkingWriter.close(events);
-		this.textWriter.close(events);
-		this.toolWriter.closeAll(events);
+		this.closeAllBlocks(events);
 		events.push({
 			type: "message_delta",
 			delta: {
@@ -47276,6 +47303,11 @@ var AnthropicStreamTranslator = class {
 			}
 		}];
 	}
+	closeAllBlocks(events) {
+		this.thinkingWriter.close(events);
+		this.textWriter.close(events);
+		this.toolWriter.closeAll(events);
+	}
 	appendMessageStart(events, chunk) {
 		if (this.state.messageStartSent) return;
 		events.push({
@@ -47940,6 +47972,106 @@ function modelSupportsOutputConfig(model) {
 	return !MODELS_REJECTING_OUTPUT_CONFIG.has(model.id);
 }
 
+//#endregion
+//#region src/lib/model-rewrite.ts
+/**
+* Unified model rewrite: user rules → built-in normalization → pass-through.
+* Call once at handler entry, before any model lookup or policy.
+*/
+function rewriteModel(modelId) {
+	const userRules = getCachedConfig().modelRewrites;
+	if (userRules) {
+		for (const rule of userRules) if (matchesGlob(rule.from, modelId)) return {
+			originalModel: modelId,
+			model: normalizeToKnownModel(rule.to) ?? rule.to
+		};
+	}
+	const normalized = normalizeToKnownModel(modelId);
+	if (normalized && normalized !== modelId) return {
+		originalModel: modelId,
+		model: normalized
+	};
+	return {
+		originalModel: modelId,
+		model: modelId
+	};
+}
+/**
+* Apply model rewrite to a mutable model field and log if changed.
+* Returns the rewrite result for downstream use.
+*/
+function applyModelRewrite(payload) {
+	const result = rewriteModel(payload.model);
+	if (result.model !== result.originalModel) {
+		consola.debug(`Model rewritten: ${result.originalModel} ~> ${result.model}`);
+		payload.model = result.model;
+	}
+	return result;
+}
+const DOT_RE = /\./g;
+/**
+* Resolve a model ID against Copilot's cached model list using
+* dash/dot equivalence. Returns the canonical ID if found.
+*/
+function normalizeToKnownModel(modelId) {
+	const models = state.cache.models?.data;
+	if (!models) return void 0;
+	if (models.some((m) => m.id === modelId)) return modelId;
+	const normalized = modelId.replace(DOT_RE, "-");
+	for (const model of models) if (model.id.replace(DOT_RE, "-") === normalized) return model.id;
+}
+const GLOB_SPECIAL_RE = /[.+^${}()|[\]\\]/g;
+const GLOB_STAR_RE = /\*/g;
+function matchesGlob(pattern, value) {
+	if (!pattern.includes("*")) return pattern === value;
+	return new RegExp(`^${pattern.replace(GLOB_SPECIAL_RE, "\\$&").replace(GLOB_STAR_RE, ".*")}$`).test(value);
+}
+/** Data-driven upgrade rules. Add new entries to extend. */
+const CONTEXT_UPGRADE_RULES = [{
+	from: "claude-opus-4.6",
+	to: "claude-opus-4.6-1m"
+}];
+/** Pre-computed set for fast model eligibility checks (avoids token estimation on non-eligible models). */
+const UPGRADE_ELIGIBLE_MODELS = new Set(CONTEXT_UPGRADE_RULES.map((r) => r.from));
+/**
+* Quick check: does this model have any context-upgrade rules?
+* Use to skip expensive token estimation for ineligible models.
+*/
+function hasContextUpgradeRule(model) {
+	return UPGRADE_ELIGIBLE_MODELS.has(model);
+}
+/** Find the upgrade rule for a model whose target exists in Copilot's model list. */
+function findUpgradeRule(model) {
+	for (const rule of CONTEXT_UPGRADE_RULES) if (model === rule.from && findModelById(rule.to)) return rule;
+}
+/**
+* Proactive: resolve the upgrade target model for a given model + token count.
+* Returns the target model ID, or undefined if no upgrade applies.
+*/
+function resolveContextUpgrade(model, estimatedTokens) {
+	const rule = findUpgradeRule(model);
+	if (rule && estimatedTokens > getContextUpgradeTokenThreshold()) return rule.to;
+}
+/**
+* Reactive: get the upgrade target for a model on context-length error.
+* Returns the target model ID, or undefined if no fallback applies.
+*/
+function getContextUpgradeTarget(model) {
+	return findUpgradeRule(model)?.to;
+}
+/** Context-length error detection with pattern matching */
+const CONTEXT_ERROR_PATTERNS = [
+	/context.length/i,
+	/too.long/i,
+	/token.*(limit|maximum|exceed)/i,
+	/(limit|maximum|exceed).*token/i
+];
+function isContextLengthError(error) {
+	if (!(error instanceof HTTPError) || error.status !== 400) return false;
+	const message = error.body?.error?.message;
+	return message ? CONTEXT_ERROR_PATTERNS.some((pattern) => pattern.test(message)) : false;
+}
+
 //#endregion
 //#region src/lib/tokenizer.ts
 const ENCODING_MAP = {
@@ -48740,7 +48872,8 @@ async function handleCompletionCore({ body, signal, headers }) {
 	const adapter = new OpenAIChatAdapter();
 	let payload = parseOpenAIChatPayload(body);
 	consola.debug("Request payload:", JSON.stringify(payload).slice(-400));
-	const originalModel = payload.model;
+	const rewrite = applyModelRewrite(payload);
+	const originalModel = rewrite.originalModel;
 	const selectedModel = findModelById(payload.model);
 	try {
 		if (selectedModel) {
@@ -48761,6 +48894,7 @@ async function handleCompletionCore({ body, signal, headers }) {
 	const plan = adapter.toCapiPlan(payload, { requestContext: readCapiRequestContext(headers) });
 	const modelMapping = {
 		originalModel,
+		rewrittenModel: rewrite.model,
 		mappedModel: plan.resolvedModel
 	};
 	const transport = new CopilotTransport(createCopilotClient());
@@ -48774,7 +48908,8 @@ async function handleCompletionCore({ body, signal, headers }) {
 //#endregion
 //#region src/routes/chat-completions/route.ts
 function createCompletionRoutes() {
-	return new Elysia().use(requestGuardPlugin).post("/chat/completions", async function* ({ body, request }) {
+	return new Elysia().use(requestGuardPlugin).post("/chat/completions", async function* ({ body, request, server }) {
+		if (hasStreamingFlag(body)) disableIdleTimeout(server, request);
 		const { result, modelMapping } = await handleCompletionCore({
 			body,
 			signal: request.signal,
@@ -48860,60 +48995,15 @@ async function handleCountTokensCore({ body, headers }) {
 	return { input_tokens: finalTokenCount };
 }
 
-//#endregion
-//#region src/lib/context-upgrade.ts
-/** Data-driven upgrade rules. Add new entries to extend. */
-const CONTEXT_UPGRADE_RULES = [{
-	from: "claude-opus-4.6",
-	to: "claude-opus-4.6-1m",
-	tokenThreshold: 19e4
-}];
-/** Pre-computed set for fast model eligibility checks (avoids token estimation on non-eligible models). */
-const UPGRADE_ELIGIBLE_MODELS = new Set(CONTEXT_UPGRADE_RULES.map((r) => r.from));
-/**
-* Quick check: does this model have any context-upgrade rules?
-* Use to skip expensive token estimation for ineligible models.
-*/
-function hasContextUpgradeRule(model) {
-	return UPGRADE_ELIGIBLE_MODELS.has(model);
-}
-/** Find the upgrade rule for a model whose target exists in Copilot's model list. */
-function findUpgradeRule(model) {
-	for (const rule of CONTEXT_UPGRADE_RULES) if (model === rule.from && findModelById(rule.to)) return rule;
-}
-/**
-* Proactive: resolve the upgrade target model for a given model + token count.
-* Returns the target model ID, or undefined if no upgrade applies.
-*/
-function resolveContextUpgrade(model, estimatedTokens) {
-	const rule = findUpgradeRule(model);
-	if (rule && estimatedTokens > rule.tokenThreshold) return rule.to;
-}
-/**
-* Reactive: get the upgrade target for a model on context-length error.
-* Returns the target model ID, or undefined if no fallback applies.
-*/
-function getContextUpgradeTarget(model) {
-	return findUpgradeRule(model)?.to;
-}
-/** Context-length error detection with pattern matching */
-const CONTEXT_ERROR_PATTERNS = [
-	/context.length/i,
-	/too.long/i,
-	/token.*(limit|maximum|exceed)/i,
-	/(limit|maximum|exceed).*token/i
-];
-function isContextLengthError(error) {
-	if (!(error instanceof HTTPError) || error.status !== 400) return false;
-	const message = error.body?.error?.message;
-	return message ? CONTEXT_ERROR_PATTERNS.some((pattern) => pattern.test(message)) : false;
-}
-
 //#endregion
 //#region src/lib/request-model-policy.ts
 const COMPACT_SYSTEM_PROMPT_START = "You are a helpful AI assistant tasked with summarizing conversations";
-function applyMessagesModelPolicy(payload) {
+function applyMessagesModelPolicy(payload, options) {
 	const originalModel = payload.model;
+	if (options?.betaUpgraded) return {
+		originalModel,
+		routedModel: originalModel
+	};
 	if (shouldContextUpgrade() && hasContextUpgradeRule(payload.model)) {
 		const contextUpgradeTarget = resolveContextUpgrade(payload.model, estimateAnthropicInputTokens(payload));
 		if (contextUpgradeTarget) {
@@ -49396,6 +49486,13 @@ function createMessagesViaChatCompletionsStrategy(transport, adapter, plan, sign
 				data: JSON.stringify(event)
 			}));
 		},
+		onStreamDone() {
+			if (!streamTranslator) return null;
+			return streamTranslator.onDone().map((event) => ({
+				event: event.type,
+				data: JSON.stringify(event)
+			}));
+		},
 		shouldBreakStream() {
 			return done;
 		},
@@ -50048,7 +50145,10 @@ function createMessagesViaResponsesStrategy(copilotClient, responsesPayload, opt
 //#endregion
 //#region src/routes/messages/strategy-registry.ts
 function selectStrategy(registry, model) {
-	for (const entry of registry) if (entry.canHandle(model)) return entry;
+	for (const entry of registry) if (entry.canHandle(model)) {
+		consola.debug(`Strategy selected: ${entry.name} for model: ${model?.id ?? "(unknown)"}`);
+		return entry;
+	}
 	return registry.at(-1);
 }
 function filterThinkingBlocksForNativeMessages(anthropicPayload) {
@@ -50142,6 +50242,30 @@ const defaultStrategyRegistry = [
 
 //#endregion
 //#region src/routes/messages/handler.ts
+const CONTEXT_BETA_RE = /^context-\d+[km]-/;
+function processAnthropicBetaHeader(rawHeader, model) {
+	if (!rawHeader) return {
+		header: void 0,
+		upgradeTarget: void 0
+	};
+	const values = rawHeader.split(",").map((v) => v.trim()).filter(Boolean);
+	let upgradeTarget;
+	const filtered = [];
+	for (const value of values) {
+		if (CONTEXT_BETA_RE.test(value)) {
+			if (!upgradeTarget && shouldContextUpgrade()) {
+				const target = getContextUpgradeTarget(model);
+				if (target) upgradeTarget = target;
+			}
+			continue;
+		}
+		filtered.push(value);
+	}
+	return {
+		header: filtered.length > 0 ? filtered.join(",") : void 0,
+		upgradeTarget
+	};
+}
 /**
 * Core handler for Anthropic messages endpoint.
 * Returns both the execution result and model mapping info.
@@ -50149,10 +50273,17 @@ const defaultStrategyRegistry = [
 async function handleMessagesCore({ body, signal, headers }) {
 	const anthropicPayload = parseAnthropicMessagesPayload(body);
 	if (consola.level >= 4) consola.debug("Anthropic request payload:", JSON.stringify(anthropicPayload));
-	const anthropicBetaHeader = headers.get("anthropic-beta") ?? void 0;
-	const modelRouting = applyMessagesModelPolicy(anthropicPayload);
+	const rewrite = applyModelRewrite(anthropicPayload);
+	const betaResult = processAnthropicBetaHeader(headers.get("anthropic-beta"), anthropicPayload.model);
+	if (betaResult.upgradeTarget) {
+		consola.debug(`Beta header context upgrade: ${anthropicPayload.model} → ${betaResult.upgradeTarget}`);
+		anthropicPayload.model = betaResult.upgradeTarget;
+	}
+	const anthropicBetaHeader = betaResult.header;
+	const modelRouting = applyMessagesModelPolicy(anthropicPayload, { betaUpgraded: !!betaResult.upgradeTarget });
 	const modelMapping = {
-		originalModel: modelRouting.originalModel,
+		originalModel: rewrite.originalModel,
+		rewrittenModel: rewrite.model,
 		mappedModel: modelRouting.routedModel
 	};
 	if (modelRouting.reason) consola.debug(`Routed anthropic request via ${modelRouting.reason}:`, `${modelRouting.originalModel} -> ${modelRouting.routedModel}`);
@@ -50186,7 +50317,8 @@ async function handleMessagesCore({ body, signal, headers }) {
 			selectedModel: retryModel,
 			upstreamSignal: retrySignal,
 			modelMapping: {
-				originalModel: modelRouting.originalModel,
+				originalModel: rewrite.originalModel,
+				rewrittenModel: rewrite.model,
 				mappedModel: upgradeTarget
 			}
 		});
@@ -50200,7 +50332,8 @@ async function handleMessagesCore({ body, signal, headers }) {
 //#endregion
 //#region src/routes/messages/route.ts
 function createMessageRoutes() {
-	return new Elysia().use(requestGuardPlugin).post("/messages", async function* ({ body, request }) {
+	return new Elysia().use(requestGuardPlugin).post("/messages", async function* ({ body, request, server }) {
+		if (hasStreamingFlag(body)) disableIdleTimeout(server, request);
 		const { result, modelMapping } = await handleMessagesCore({
 			body,
 			signal: request.signal,
@@ -50305,6 +50438,7 @@ const HTTP_URL_RE = /^https?:\/\//i;
 */
 async function handleResponsesCore({ body, signal, headers }) {
 	const payload = parseResponsesPayload(body);
+	const rewrite = applyModelRewrite(payload);
 	applyResponsesToolTransforms(payload);
 	applyResponsesInputPolicies(payload);
 	compactInputByLatestCompaction(payload);
@@ -50314,12 +50448,19 @@ async function handleResponsesCore({ body, signal, headers }) {
 	applyContextManagement(payload, selectedModel.capabilities.limits.max_prompt_tokens);
 	const { vision, initiator } = getResponsesRequestOptions(payload);
 	const upstreamSignal = createUpstreamSignalFromConfig(signal);
-	return runStrategy(createResponsesPassthroughStrategy(createCopilotClient(), payload, {
-		vision,
-		initiator,
-		requestContext: readCapiRequestContext(headers),
-		signal: upstreamSignal.signal
-	}), upstreamSignal);
+	return {
+		result: await runStrategy(createResponsesPassthroughStrategy(createCopilotClient(), payload, {
+			vision,
+			initiator,
+			requestContext: readCapiRequestContext(headers),
+			signal: upstreamSignal.signal
+		}), upstreamSignal),
+		modelMapping: {
+			originalModel: rewrite.originalModel,
+			rewrittenModel: rewrite.model,
+			mappedModel: payload.model
+		}
+	};
 }
 function applyResponsesToolTransforms(payload) {
 	applyFunctionApplyPatch(payload);
@@ -50454,12 +50595,14 @@ function parseBooleanParam(value) {
 //#endregion
 //#region src/routes/responses/route.ts
 function createResponsesRoutes() {
-	return new Elysia().use(requestGuardPlugin).post("/responses", async function* ({ body, request }) {
-		const result = await handleResponsesCore({
+	return new Elysia().use(requestGuardPlugin).post("/responses", async function* ({ body, request, server }) {
+		if (hasStreamingFlag(body)) disableIdleTimeout(server, request);
+		const { result, modelMapping } = await handleResponsesCore({
 			body,
 			signal: request.signal,
 			headers: request.headers
 		});
+		if (modelMapping) setRequestModelMapping(request, modelMapping);
 		if (result.kind === "json") return result.data;
 		yield* sseAdapter(result.generator);
 	}, { guarded: true }).post("/responses/input_tokens", async ({ body, request }) => {
@@ -50475,7 +50618,8 @@ function createResponsesRoutes() {
 			headers: request.headers,
 			signal: request.signal
 		});
-	}).get("/responses/:responseId", async ({ params, request }) => {
+	}).get("/responses/:responseId", async ({ params, request, server }) => {
+		if (hasStreamingResponsesQuery(request)) disableIdleTimeout(server, request);
 		return handleRetrieveResponseCore({
 			params,
 			url: request.url,