copilot-relay 0.5.2 → 0.5.3

package/dist/main.js

@@ -5401,12 +5401,14 @@ function normalizeResponsesToolSchema(schema) {
 
 //#endregion
 //#region src/copilot/responses.ts
+const minResponsesOutputTokens = 16;
+const responsesOutputTokens = (maxTokens) => typeof maxTokens === "number" && Number.isInteger(maxTokens) && maxTokens > 0 && maxTokens < minResponsesOutputTokens ? minResponsesOutputTokens : maxTokens;
 function buildResponsesRequestPayload(payload, reasoningEffort) {
 	return {
 		model: payload.model,
 		input: translateMessagesToResponsesInput(payload.messages),
 		stream: payload.stream,
-		max_output_tokens: payload.max_tokens,
+		max_output_tokens: responsesOutputTokens(payload.max_tokens),
 		temperature: payload.temperature,
 		top_p: payload.top_p,
 		user: sanitizeUserIdentifier(payload.user),
@@ -10653,11 +10655,35 @@ const claudeAppDataDir = ({ platform, env, home } = currentPlatform()) => {
 	return path.posix.join(configHome, "Claude-3p");
 };
 const standardDataDir = (dataDir) => dataDir.replace(/-3p$/, "");
+const otherVendorMarker = /ark-code|astron|command-r|deepseek|doubao|gemini|gemma|glm|gpt|grok|hermes|hy3|kimi|lfm|\bling\b|llama|longcat|mimo|minimax|mistral|mixtral|moonshot|nemotron|openai|phi-|qianfan|qwen|tc-code|\bunic\b|yi-|stepfun|step-3|seed-|bytedance|hunyuan|granite|amazon\.nova|nova-|devstral|ministral|ernie|codex|arcee|trinity|abab|phi\d|\bk2\.|\bm2\.|jamba|arctic|solar|mercury|zamba|kat-coder|\bds-|dpsk/;
+const looksAnthropic = (id) => {
+	const lower = id.toLowerCase();
+	return !otherVendorMarker.test(lower) && [
+		"claude",
+		"anthropic",
+		"opus",
+		"sonnet",
+		"haiku",
+		"fable",
+		"mythos"
+	].some((word) => lower.includes(word));
+};
+const routesToTier = (name, tier) => name.toLowerCase().includes("opus") === (tier === "opus");
+/**
+* The name the app sends for a model: its own ID when the app accepts it and the relay routes that
+* ID to the model's own slot, otherwise a Claude-style route for its tier, such as
+* claude-sonnet-relay. Each tier therefore has its own name. The relay sends every model name without "opus"
+* to gptModel and every one with it to opusModel, so the route reaches the same upstream model. The
+* route names no real Claude model, so the app and its Claude Code take none of that model's
+* built-in context window or tuning; supports1m on the entry says what the window is.
+*/
+const appModelName = (model) => looksAnthropic(model.id) && routesToTier(model.id, model.tier) ? model.id : `claude-${model.tier}-relay`;
 const inferenceModel = (model) => {
 	const promptTokens = model.capabilities?.limits?.max_prompt_tokens ?? 0;
-	const label = model.capabilities?.name?.trim();
+	const name = appModelName(model);
+	const label = model.capabilities?.name?.trim() || (name === model.id ? void 0 : model.id);
 	return {
-		name: model.id,
+		name,
 		...label ? { labelOverride: label } : {},
 		...promptTokens >= 1e6 ? { supports1m: true } : {},
 		anthropicFamilyTier: model.tier
@@ -10730,7 +10756,8 @@ const applyClaudeAppSetup = async (input) => {
 		inferenceGatewayBaseUrl: input.baseUrl,
 		inferenceGatewayApiKey: input.apiKey || existingKey || "copilot-relay",
 		inferenceCredentialKind: "static",
-		inferenceModels: input.models.map(inferenceModel)
+		inferenceModels: input.models.map(inferenceModel),
+		modelDiscoveryEnabled: false
 	};
 	const desktopSnapshot = await readFileSnapshot(path.join(dataDir, "claude_desktop_config.json"));
 	const desktop = {