anygate 0.5.5 → 0.5.7

package/dist/{chunk-CH5IEXJN.js → chunk-BCPX3QLK.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   getTemplateById
-} from "./chunk-YYSUTRMV.js";
+} from "./chunk-VKROC37K.js";
 
 // src/core/constants.ts
 import { homedir } from "os";
@@ -10,7 +10,7 @@ import { join } from "path";
 // package.json
 var package_default = {
   name: "anygate",
-  version: "0.5.5",
+  version: "0.5.7",
   publishConfig: {
     access: "public"
   },
@@ -3780,6 +3780,41 @@ function findModelsDevModel(providerId, modelId, cache = loadModelsDevCache()) {
   }
   return null;
 }
+var MULTIMODAL_FAMILIES = [
+  /nvidia.*nemotron/i,
+  /^gpt-/i,
+  /^o[0-9]/i,
+  // o1, o3, o4, o-series
+  /gpt-5/i,
+  /gemini/i,
+  /claude/i,
+  /sonnet/i,
+  /opus/i,
+  /haiku/i,
+  /llama.*vision/i,
+  /(vision|multimodal)/i,
+  /qwen.*vl/i,
+  /qwen2?-vl/i,
+  /deepseek.*vl/i,
+  /mistral.*(vision|pixtral)/i,
+  /pixtral/i,
+  /grok.*vision/i,
+  /(command|cohere).*vision/i
+];
+function familyMatchesMultimodal(family, modelId) {
+  const hay = `${family} ${modelId}`;
+  return MULTIMODAL_FAMILIES.some((re) => re.test(hay));
+}
+function resolveInputTypes(family, providerId, modelId, cache = loadModelsDevCache()) {
+  const entry = findModelsDevModel(providerId, modelId, cache);
+  const baseInput = entry?.modalities?.input && entry.modalities.input.length > 0 ? [...entry.modalities.input] : null;
+  if (baseInput && baseInput.every((t) => t === "text") && !familyMatchesMultimodal(family, modelId)) {
+    return ["text"];
+  }
+  const result = new Set(baseInput ?? ["text"]);
+  if (familyMatchesMultimodal(family, modelId)) result.add("image");
+  return [...result];
+}
 function shouldHideByModelsDevCapabilities(entry) {
   const output = entry.modalities?.output;
   if (output && output.length > 0 && !output.includes("text")) return true;
@@ -3981,7 +4016,7 @@ function maskGatewayModelId(aliasId) {
 // src/gateway/models.ts
 var CREATED_AT_ISO = "2025-01-01T00:00:00Z";
 var CREATED_AT_UNIX = 1735689600;
-function formatAnthropicModelEntry(id, displayName, contextWindow) {
+function formatAnthropicModelEntry(id, displayName, contextWindow, inputTypes) {
   const maxInput = resolveContextWindow(id, contextWindow);
   return {
     id,
@@ -3989,12 +4024,13 @@ function formatAnthropicModelEntry(id, displayName, contextWindow) {
     display_name: displayName,
     created_at: CREATED_AT_ISO,
     context_window: maxInput,
-    max_input_tokens: maxInput
+    max_input_tokens: maxInput,
+    input_types: inputTypes ?? ["text"]
   };
 }
 function formatAnthropicModelList(entries) {
   return {
-    data: entries.map((entry) => formatAnthropicModelEntry(entry.id, entry.name, entry.contextWindow)),
+    data: entries.map((entry) => formatAnthropicModelEntry(entry.id, entry.name, entry.contextWindow, entry.inputTypes)),
     has_more: false,
     first_id: entries[0]?.id ?? null,
     last_id: entries.at(-1)?.id ?? null
@@ -4022,7 +4058,8 @@ function formatGatewayAnthropicModels(models, opts) {
     models.map((model) => ({
       id: exposedGatewayAliasId(model, opts),
       name: gatewayDisplayName(model, opts),
-      contextWindow: model.contextWindow
+      contextWindow: model.contextWindow,
+      inputTypes: model.supportedParameters?.includes("image") ? ["text", "image"] : ["text"]
     }))
   );
 }
@@ -5103,6 +5140,94 @@ function resolveUpstreamTools(tools, messages) {
   return upstream;
 }
 
+// src/gateway/context-fit.ts
+var CHARS_PER_TOKEN = 4;
+var RESERVE_TOKENS = 256;
+function estimateTokens(text4) {
+  return Math.ceil(text4.length / CHARS_PER_TOKEN);
+}
+function messageTokens(msg) {
+  if (typeof msg.content === "string") return estimateTokens(msg.content);
+  let total = 0;
+  for (const block of msg.content) {
+    if (typeof block.text === "string") {
+      total += estimateTokens(block.text);
+    }
+    if (block.source?.data) {
+      total += Math.max(512, estimateTokens(block.source.data));
+    }
+    if (typeof block.input === "object" && block.input !== null) {
+      try {
+        total += estimateTokens(JSON.stringify(block.input));
+      } catch {
+      }
+    }
+  }
+  return total;
+}
+function estimateContextTokens(system, messages) {
+  let total = estimateTokens(system ?? "");
+  for (const msg of messages) total += messageTokens(msg);
+  return total;
+}
+function fitContextWindow(messages, system, contextWindow, maxOutputTokens) {
+  const systemTokens = estimateTokens(system ?? "");
+  const budget = contextWindow - maxOutputTokens - RESERVE_TOKENS;
+  if (budget <= 0) {
+    return { system, messages, trimmed: false, dropped: 0 };
+  }
+  const usable = budget - systemTokens;
+  if (usable <= 0) {
+    return { system, messages, trimmed: false, dropped: 0 };
+  }
+  const kept = [];
+  let used = 0;
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const msg = messages[i];
+    const cost = messageTokens(msg);
+    if (used + cost > usable && kept.length > 0) break;
+    kept.unshift(msg);
+    used += cost;
+  }
+  const toolUseIds = /* @__PURE__ */ new Set();
+  const toolResultIds = /* @__PURE__ */ new Set();
+  for (const msg of kept) {
+    if (!Array.isArray(msg.content)) continue;
+    for (const b of msg.content) {
+      if (b.type === "tool_use" && b.id) toolUseIds.add(b.id);
+      if (b.type === "tool_result" && b.tool_use_id) toolResultIds.add(b.tool_use_id);
+    }
+  }
+  const orphanResult = kept.filter((msg) => {
+    if (!Array.isArray(msg.content)) return false;
+    return msg.content.some((b) => b.type === "tool_result" && b.tool_use_id && !toolUseIds.has(b.tool_use_id));
+  });
+  const orphanUse = kept.filter((msg) => {
+    if (!Array.isArray(msg.content)) return false;
+    return msg.content.some((b) => b.type === "tool_use" && b.id && !toolResultIds.has(b.id));
+  });
+  let trimmedMessages = kept;
+  if (orphanResult.length || orphanUse.length) {
+    const dropIds = /* @__PURE__ */ new Set();
+    for (const msg of [...orphanResult, ...orphanUse]) {
+      for (const b of msg.content) {
+        if (b.id) dropIds.add(b.id);
+        if (b.tool_use_id) dropIds.add(b.tool_use_id);
+      }
+    }
+    trimmedMessages = trimmedMessages.filter(
+      (msg) => !msg.content || !Array.isArray(msg.content) || !msg.content.some((b) => b.id && dropIds.has(b.id) || b.tool_use_id && dropIds.has(b.tool_use_id))
+    );
+  }
+  const dropped = messages.length - trimmedMessages.length;
+  return {
+    system,
+    messages: trimmedMessages,
+    trimmed: dropped > 0,
+    dropped
+  };
+}
+
 // src/gateway/sdk-adapter.ts
 function anthropicEffortFromRequest(body) {
   const effort = body.output_config?.effort;
@@ -5240,11 +5365,32 @@ function translateToolChoice(tc) {
   return void 0;
 }
 function translateRequest2(body, npm, options) {
-  const messages = body.messages ?? [];
+  let messages = body.messages ?? [];
   annotateToolNames(messages);
   const baseSystem = systemToString(body.system);
   const inlineParts = inlineSystemText(messages);
   const systemText = [baseSystem, ...inlineParts].filter((s) => s && s.trim()).join("\n\n") || (options?.openAiOAuth ? "You are a coding assistant." : void 0);
+  let trimmedSystem = systemText;
+  let maxOutput = options?.openAiOAuth ? void 0 : body.max_tokens;
+  if (options?.contextWindow && options.contextWindow > 0) {
+    const { system: fittedSystem, messages: fittedMessages, trimmed } = fitContextWindow(
+      messages,
+      systemText,
+      options.contextWindow,
+      typeof maxOutput === "number" ? maxOutput : 0
+    );
+    if (trimmed) {
+      messages = fittedMessages;
+      annotateToolNames(messages);
+      const fittedInline = inlineSystemText(messages);
+      trimmedSystem = [fittedSystem, ...fittedInline].filter((s) => s && s.trim()).join("\n\n") || (options?.openAiOAuth ? "You are a coding assistant." : void 0);
+      if (typeof maxOutput === "number") {
+        const fittedInputTokens = estimateContextTokens(fittedSystem ?? "", fittedMessages);
+        const headroom = options.contextWindow - fittedInputTokens - 256;
+        if (headroom > 0 && maxOutput > headroom) maxOutput = headroom;
+      }
+    }
+  }
   let upstreamTools = resolveUpstreamTools(
     body.tools,
     messages
@@ -5263,11 +5409,11 @@ function translateRequest2(body, npm, options) {
     });
   }
   return {
-    system: options?.openAiOAuth ? void 0 : systemText,
+    system: options?.openAiOAuth ? void 0 : trimmedSystem,
     messages: translateMessages(messages, npm),
     tools: translateTools3(upstreamTools.length ? upstreamTools : void 0),
     toolChoice: translateToolChoice(body.tool_choice),
-    maxOutputTokens: options?.openAiOAuth ? void 0 : body.max_tokens,
+    maxOutputTokens: options?.openAiOAuth ? void 0 : maxOutput,
     temperature: body.temperature,
     providerOptions
   };
@@ -5698,6 +5844,10 @@ function makeProxyLog(debug, logPath) {
     appendSecureLog(path, line);
   };
 }
+var quietErrorLogPath = getProxyDebugLogPath();
+function quietErrorLog(line) {
+  appendSecureLog(quietErrorLogPath, `[quiet-error] ${line}`);
+}
 function anthropicError(res, status, message) {
   sendJson(res, status, {
     type: "error",
@@ -5707,7 +5857,8 @@ function anthropicError(res, status, message) {
 function aliasModelId(realId, providerId) {
   if (realId.startsWith("claude-")) return realId;
   const sanitized = providerId.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-+|-+$/g, "");
-  return `anthropic-${sanitized}__${realId}`;
+  const safeRealId = realId.replace(/[\/\s:()]+/g, "-");
+  return `anthropic-${sanitized}__${safeRealId}`;
 }
 function lookupRoute(byAlias, id) {
   for (const key of routeLookupIds(id)) {
@@ -5735,7 +5886,7 @@ function startProxyCatalog(routes, defaultAliasId, debug = false) {
   process.on("uncaughtException", onException);
   const modelsPayload = JSON.stringify(
     formatAnthropicModelList(
-      routes.map((r) => ({ id: r.aliasId, name: r.displayName, contextWindow: r.contextWindow }))
+      routes.map((r) => ({ id: r.aliasId, name: r.displayName, contextWindow: r.contextWindow, inputTypes: r.inputTypes }))
     )
   );
   const server = createServer(async (req, res) => {
@@ -5752,8 +5903,9 @@ function startProxyCatalog(routes, defaultAliasId, debug = false) {
         const route = lookupRoute(byAlias, id);
         if (route) {
           res.writeHead(200, { "Content-Type": "application/json" });
-          res.end(JSON.stringify(formatAnthropicModelEntry(route.aliasId, route.displayName, route.contextWindow)));
+          res.end(JSON.stringify(formatAnthropicModelEntry(route.aliasId, route.displayName, route.contextWindow, route.inputTypes)));
         } else {
+          quietErrorLog(`GET /v1/models/${id} - model not found`);
           res.writeHead(404, { "Content-Type": "application/json" });
           res.end(JSON.stringify({ error: { type: "not_found_error", message: `Model '${id}' not found` } }));
         }
@@ -5780,6 +5932,9 @@ function startProxyCatalog(routes, defaultAliasId, debug = false) {
       const originalModel = anthropicBody.model;
       const clientWantsStream = Boolean(anthropicBody.stream);
       const route = lookupRoute(byAlias, originalModel) ?? defaultRoute;
+      if (route === defaultRoute && originalModel !== defaultRoute.aliasId) {
+        quietErrorLog(`POST /v1/messages - model alias '${originalModel}' not found, falling back to default route '${defaultRoute.aliasId}'`);
+      }
       const apiKey = route.apiKey;
       const upstreamUrl = route.upstreamUrl;
       plog(
@@ -5845,7 +6000,8 @@ function startProxyCatalog(routes, defaultAliasId, debug = false) {
             reasoning: route.reasoning,
             interleavedReasoningField: route.interleavedReasoningField,
             upstreamModelId: route.realModelId
-          }
+          },
+          contextWindow: route.contextWindow
         });
         plog(
           () => `sdk: npm=${route.npm} model=${route.realModelId}, stream=${clientWantsStream}, tools=${anthropicBody.tools?.length ?? 0}, msgs=${params.messages.length}`
@@ -6865,7 +7021,8 @@ function localModelToRoute(lp, model) {
     reasoning: model.reasoning,
     interleavedReasoningField: model.interleavedReasoningField,
     useResponsesLite: model.useResponsesLite,
-    preferWebSockets: model.preferWebSockets
+    preferWebSockets: model.preferWebSockets,
+    inputTypes: resolveInputTypes(model.family, lp.id, model.id)
   };
 }
 function makeRouteResolver(localProviders) {
@@ -7541,7 +7698,8 @@ async function handleAnthropicMessages(req, res, options, modelCache, plog) {
         interleavedReasoningField: model.interleavedReasoningField,
         upstreamModelId: upstreamModelId(model)
       },
-      maxTools: npmMaxTools
+      maxTools: npmMaxTools,
+      contextWindow: model.contextWindow ?? resolveContextWindow(model.id)
     });
     const clientWantsStream = Boolean(body.stream);
     const responseModelId = getResponseModelId(body.model, model, options);
@@ -7583,7 +7741,14 @@ async function handleAnthropicMessages(req, res, options, modelCache, plog) {
       if (!res.headersSent) {
         const status = upstreamHttpStatus(err);
         sendJson(res, status === 500 ? 502 : status, { error: { message } });
-      } else res.end();
+      } else {
+        const errorType = anthropicErrorType(upstreamHttpStatus(err));
+        res.write(`event: error
+data: ${JSON.stringify({ type: "error", error: { type: errorType, message } })}
+
+`);
+        res.end();
+      }
     }
     return;
   }
@@ -7656,7 +7821,14 @@ async function handleOpenAIChatCompletions(req, res, options, modelCache, plog)
     if (!res.headersSent) {
       const status = upstreamHttpStatus(err);
       sendJson(res, status === 500 ? 502 : status, { error: { message } });
-    } else res.end();
+    } else {
+      const errorType = anthropicErrorType(upstreamHttpStatus(err));
+      res.write(`event: error
+data: ${JSON.stringify({ type: "error", error: { type: errorType, message } })}
+
+`);
+      res.end();
+    }
   }
 }
 function lookupModel(res, catalog, modelId) {
@@ -10745,6 +10917,7 @@ export {
   isFreeStatus,
   freeStatusLabel,
   refreshModelsDevCacheAsync,
+  resolveInputTypes,
   shouldHideModel,
   cachedModelToLocal,
   readBody,
@@ -10782,6 +10955,7 @@ export {
   providersForPicker,
   formatRegistryAuthLabel,
   resolveProvidersForDisplay,
+  localProvidersToServerModels,
   makeRouteResolver,
   buildCatalogRoutes,
   findOpencodeBinary,
@@ -10842,4 +11016,4 @@ export {
   quitClaudeAppGracefully,
   launchOrRestartClaudeApp
 };
-//# sourceMappingURL=chunk-CH5IEXJN.js.map
+//# sourceMappingURL=chunk-BCPX3QLK.js.map