@axlsdk/axl 0.22.0 → 0.22.2

package/README.md

@@ -67,7 +67,7 @@ agent({ model: 'deepseek:deepseek-reasoner' });           // DEEPSEEK_API_KEY
 agent({ model: 'ollama:llama3' });                        // local — no key, $0
 ```
 
-Presets: `openrouter`, `azure`, `xai`, `deepseek`, `mistral`, `groq`, `bedrock`, and self-hosted `ollama` / `vllm` / `lmstudio` / `llamacpp` / `sglang`. The unified `effort` knob and per-call cost tracking work across them. Build your own by cloning a `ProviderProfile`. See [docs/providers.md](../../docs/providers.md#openai-compatible-providers--presets).
+Presets: `openrouter`, `azure`, `xai`, `deepseek`, `mistral`, `groq`, `bedrock`, and self-hosted `ollama` / `vllm` / `lmstudio` / `llamacpp` / `sglang`. The unified `effort` knob and per-call cost tracking work across them. OpenAI Responses, Anthropic, Google, and OpenRouter accept image input without hard-coding a model catalog; the selected upstream model remains authoritative. Completed-file transcription is a separate, narrower contract. Build your own compatible provider by cloning a `ProviderProfile`; see [provider details](../../docs/providers.md) and [multimodal input](../../docs/multimodal-input.md).
 
 Each provider also accepts an opt-in `rateLimit` (`{ maxConcurrent?, minIntervalMs?, acquireTimeoutMs? }`) for proactive client-side pacing on top of the automatic 429/503/529 backoff — useful when a large fan-out (e.g. an eval) shares one API key. It caps in-flight request concurrency (not token throughput) for that provider's chat calls. See [Providers → Rate limiting](../../docs/providers.md#rate-limiting-opt-in).
 
@@ -792,10 +792,12 @@ See [docs/providers.md](../../docs/providers.md) for the full model list includi
 
 Use `ctx.transcribe()` for finite recordings, then explicitly pass
 `Transcript.text` to an agent if analysis is needed. It is separate from chat
-history and chat providers. B1 supports exact OpenAI, Gemini, and OpenRouter
-transcription URIs; see [Multimodal input](../../docs/multimodal-input.md) for
-the 25 MiB inline bound, source types, options, transcript redaction, temporary
-Gemini Files cleanup, and live-verification commands.
+history and chat providers. OpenAI and Gemini use their documented model
+contracts; any nonblank `openrouter-transcription:<vendor/model>` URI can
+dispatch supported bytes/base64 to OpenRouter, where the selected model/route
+decides endpoint compatibility. See [Multimodal input](../../docs/multimodal-input.md)
+for the 25 MiB inline bound, source types, options, transcript redaction,
+temporary Gemini Files cleanup, and live-verification commands.
 
 ## License
 

package/dist/index.cjs

@@ -1277,13 +1277,12 @@ function unsafeTransportError(surface, remediation, exposedData) {
 }
 
 // src/providers/openai-compatible.ts
-var OPENROUTER_IMAGE_MODEL = "openai/gpt-4o-mini";
 function compatibleImageBase64(source) {
   return source.type === "base64" ? source.data : Buffer.from(source.data.buffer, source.data.byteOffset, source.data.byteLength).toString(
     "base64"
   );
 }
-function openRouterImageParts(parts) {
+function openRouterImageParts(parts, model) {
   const content = [];
   for (const part of parts) {
     if (part.type === "text") {
@@ -1294,7 +1293,7 @@ function openRouterImageParts(parts) {
     if (source.type === "provider-file") {
       throw new UnsupportedModelInputError({
         provider: "openrouter",
-        model: OPENROUTER_IMAGE_MODEL,
+        model,
         modality: "image",
         source: "provider-file"
       });
@@ -1425,10 +1424,11 @@ function normalizeMessageContent(content) {
 var OpenAICompatibleProvider = class {
   name;
   inputCapabilities(model) {
-    return this.name === "openrouter" && model === OPENROUTER_IMAGE_MODEL ? { image: { sources: ["url", "bytes", "base64"] } } : {};
+    return this.name === "openrouter" && model.trim().length > 0 ? { image: { sources: ["url", "bytes", "base64"] } } : {};
   }
   validateInput(request) {
-    const effectiveModel = typeof request.providerOptions?.model === "string" ? request.providerOptions.model : request.model;
+    const modelOverride = request.providerOptions?.model;
+    const effectiveModel = typeof modelOverride === "string" ? modelOverride : request.model;
     const fail = (source, feature) => {
       throw new UnsupportedModelInputError({
         provider: this.name,
@@ -1438,13 +1438,15 @@ var OpenAICompatibleProvider = class {
         ...feature ? { feature } : {}
       });
     };
-    if (this.name !== "openrouter" || effectiveModel !== OPENROUTER_IMAGE_MODEL) {
+    if (request.providerOptions && "model" in request.providerOptions && (typeof modelOverride !== "string" || modelOverride.trim().length === 0)) {
+      fail(void 0, "invalid model providerOptions");
+    }
+    if (this.name !== "openrouter" || effectiveModel.trim().length === 0) {
       fail(void 0, "image input for this model");
     }
     if (request.providerOptions && "messages" in request.providerOptions) {
       fail(void 0, "raw messages providerOptions");
     }
-    if (request.hasTools) fail(void 0, "tools with image input");
     for (const message of request.history) {
       if (!Array.isArray(message.content)) continue;
       if (message.role !== "user") fail(void 0, "rich non-user history");
@@ -1632,7 +1634,7 @@ var OpenAICompatibleProvider = class {
   formatMessage(msg, model) {
     const out = {
       role: this.profile.roleFor ? this.profile.roleFor(msg.role, model) : msg.role,
-      content: Array.isArray(msg.content) && this.name === "openrouter" && model === OPENROUTER_IMAGE_MODEL ? openRouterImageParts(msg.content) : msg.content
+      content: Array.isArray(msg.content) && this.name === "openrouter" ? openRouterImageParts(msg.content, model) : msg.content
     };
     if (msg.name && (this.profile.capabilities?.emitsMessageName ?? true)) out.name = msg.name;
     if (msg.tool_calls) out.tool_calls = msg.tool_calls;
@@ -2379,13 +2381,6 @@ var OpenAIProvider = class extends OpenAICompatibleProvider {
 };
 
 // src/providers/openai-responses.ts
-var OPENAI_RESPONSES_IMAGE_MODELS = /* @__PURE__ */ new Set([
-  "gpt-4o",
-  "gpt-4o-2024-08-06",
-  "gpt-4o-2024-11-20",
-  "gpt-4o-mini",
-  "gpt-4o-mini-2024-07-18"
-]);
 function base64FromSource(source) {
   return source.type === "base64" ? source.data : Buffer.from(source.data.buffer, source.data.byteOffset, source.data.byteLength).toString(
     "base64"
@@ -2424,10 +2419,11 @@ function responseImageParts(parts) {
 var OpenAIResponsesProvider = class {
   name = "openai-responses";
   inputCapabilities(model) {
-    return OPENAI_RESPONSES_IMAGE_MODELS.has(model) ? { image: { sources: ["url", "bytes", "base64", "provider-file"] } } : {};
+    return model.trim().length > 0 ? { image: { sources: ["url", "bytes", "base64", "provider-file"] } } : {};
   }
   validateInput(request) {
-    const effectiveModel = typeof request.providerOptions?.model === "string" ? request.providerOptions.model : request.model;
+    const modelOverride = request.providerOptions?.model;
+    const effectiveModel = typeof modelOverride === "string" ? modelOverride : request.model;
     const fail = (source, feature) => {
       throw new UnsupportedModelInputError({
         provider: this.name,
@@ -2437,7 +2433,10 @@ var OpenAIResponsesProvider = class {
         ...feature ? { feature } : {}
       });
     };
-    if (!OPENAI_RESPONSES_IMAGE_MODELS.has(effectiveModel)) {
+    if (request.providerOptions && "model" in request.providerOptions && (typeof modelOverride !== "string" || modelOverride.trim().length === 0)) {
+      fail(void 0, "invalid model providerOptions");
+    }
+    if (effectiveModel.trim().length === 0) {
       fail(void 0, "image input for this model");
     }
     if (request.providerOptions && "input" in request.providerOptions) {
@@ -2565,7 +2564,7 @@ var OpenAIResponsesProvider = class {
     const resolved = resolveThinkingOptions(options);
     const { includeThoughts } = resolved;
     const wireEffort = resolveOpenAIReasoningEffort(effectiveModel, resolved);
-    const stripTemp = oSeries || reasoningCapable && wireEffort !== void 0;
+    const stripTemp = oSeries || supportsMaxReasoningEffort(effectiveModel) || reasoningCapable && wireEffort !== void 0;
     const systemMessages = messages.filter((m) => m.role === "system");
     const nonSystemMessages = messages.filter((m) => m.role !== "system");
     const body = {
@@ -2913,7 +2912,6 @@ function anthropicImageBlocks(parts) {
 }
 var ANTHROPIC_API_VERSION = "2023-06-01";
 var ANTHROPIC_FILES_BETA = "files-api-2025-04-14";
-var ANTHROPIC_IMAGE_MODELS = /* @__PURE__ */ new Set(["claude-sonnet-4-5", "claude-opus-4-8"]);
 function hasAnthropicProviderFileImage(messages) {
   return messages.some(
     (message) => Array.isArray(message.content) && message.content.some(
@@ -3032,6 +3030,28 @@ var ANTHROPIC_RATES = {
 function isValidTokenCount(value) {
   return value !== void 0 && Number.isSafeInteger(value) && value >= 0;
 }
+function anthropicStreamErrorStatus(type) {
+  switch (type) {
+    case "invalid_request_error":
+      return 400;
+    case "authentication_error":
+      return 401;
+    case "permission_error":
+      return 403;
+    case "not_found_error":
+      return 404;
+    case "request_too_large":
+      return 413;
+    case "rate_limit_error":
+      return 429;
+    case "api_error":
+      return 500;
+    case "overloaded_error":
+      return 529;
+    default:
+      return 0;
+  }
+}
 function estimateAnthropicCost(model, usage, now = /* @__PURE__ */ new Date()) {
   if (!model || !isValidTokenCount(usage.inputTokens) || !isValidTokenCount(usage.outputTokens)) {
     return void 0;
@@ -3281,10 +3301,11 @@ function mergeAnthropicUsage(initial, update) {
 var AnthropicProvider = class {
   name = "anthropic";
   inputCapabilities(model) {
-    return ANTHROPIC_IMAGE_MODELS.has(model) ? { image: { sources: ["url", "bytes", "base64", "provider-file"] } } : {};
+    return model.trim().length > 0 ? { image: { sources: ["url", "bytes", "base64", "provider-file"] } } : {};
   }
   validateInput(request) {
-    const effectiveModel = typeof request.providerOptions?.model === "string" ? request.providerOptions.model : request.model;
+    const modelOverride = request.providerOptions?.model;
+    const effectiveModel = typeof modelOverride === "string" ? modelOverride : request.model;
     const fail = (source, feature) => {
       throw new UnsupportedModelInputError({
         provider: this.name,
@@ -3294,7 +3315,10 @@ var AnthropicProvider = class {
         ...feature ? { feature } : {}
       });
     };
-    if (!ANTHROPIC_IMAGE_MODELS.has(effectiveModel)) {
+    if (request.providerOptions && "model" in request.providerOptions && (typeof modelOverride !== "string" || modelOverride.trim().length === 0)) {
+      fail(void 0, "invalid model providerOptions");
+    }
+    if (effectiveModel.trim().length === 0) {
       fail(void 0, "image input for this model");
     }
     if (request.providerOptions && "messages" in request.providerOptions) {
@@ -3419,7 +3443,7 @@ var AnthropicProvider = class {
     if (!res.body) {
       throw new Error("Anthropic stream response has no body");
     }
-    yield* this.parseSSEStream(res.body, pricingContext);
+    yield* this.parseSSEStream(res.body, pricingContext, res.headers);
   }
   // ---------------------------------------------------------------------------
   // Internal: request building
@@ -3662,7 +3686,7 @@ ${jsonInstruction}` : jsonInstruction;
   // ---------------------------------------------------------------------------
   // Internal: SSE stream parsing
   // ---------------------------------------------------------------------------
-  async *parseSSEStream(body, pricingContext) {
+  async *parseSSEStream(body, pricingContext, responseHeaders) {
     const reader = body.getReader();
     const decoder = new TextDecoder();
     let buffer = "";
@@ -3708,6 +3732,15 @@ ${jsonInstruction}` : jsonInstruction;
             continue;
           }
           switch (event.type) {
+            case "error": {
+              const status = anthropicStreamErrorStatus(event.error?.type);
+              throw buildProviderError({
+                provider: this.name,
+                status,
+                headers: responseHeaders,
+                message: event.error?.message ?? "Anthropic stream failed"
+              });
+            }
             case "content_block_start": {
               const block = event.content_block;
               if (block?.type === "tool_use") {
@@ -3959,7 +3992,6 @@ var GEMINI_PRICING = {
   "gemini-3.5-flash-lite": { input: 3e-7, cached: 3e-8, output: 25e-7 },
   "gemini-3.6-flash": { input: 15e-7, cached: 15e-8, output: 75e-7 }
 };
-var GEMINI_INTERACTIONS_IMAGE_MODELS = /* @__PURE__ */ new Set(["gemini-3.7-flash"]);
 function isValidTokenCount2(value) {
   return typeof value === "number" && Number.isSafeInteger(value) && value >= 0;
 }
@@ -3969,6 +4001,55 @@ function isStandardGeminiRequestTier(value) {
 function isDefinitiveStandardGeminiResponseTier(value) {
   return value === "SERVICE_TIER_UNSPECIFIED" || value === "SERVICE_TIER_STANDARD" || value === "standard";
 }
+function geminiInteractionErrorStatus(code) {
+  if (typeof code === "number" && Number.isInteger(code) && code >= 100 && code <= 599) {
+    return code;
+  }
+  switch (code) {
+    case "invalid_request":
+    case "invalid_argument":
+    case "parameter_unknown":
+    case "failed_precondition":
+      return 400;
+    case "authentication":
+    case "unauthenticated":
+      return 401;
+    case "permission_denied":
+      return 403;
+    case "not_found":
+    case "model_not_found":
+      return 404;
+    case "request_timeout":
+      return 408;
+    case "conflict":
+    case "already_exists":
+      return 409;
+    case "request_too_large":
+      return 413;
+    case "out_of_range":
+      return 416;
+    case "quota_exceeded":
+    case "resource_exhausted":
+    case "rate_limit_exceeded":
+      return 429;
+    case "cancelled":
+      return 499;
+    case "api_error":
+    case "internal":
+    case "internal_server_error":
+      return 500;
+    case "bad_gateway":
+      return 502;
+    case "unavailable":
+    case "service_unavailable":
+      return 503;
+    case "deadline_exceeded":
+    case "gateway_timeout":
+      return 504;
+    default:
+      return void 0;
+  }
+}
 function isEligibleGeminiPricing(context, hasDefinitiveStandardResponseTier, hasInvalidResponseTier) {
   return context.eligibleRequest && isStandardGeminiRequestTier(context.serviceTier) && hasDefinitiveStandardResponseTier && !hasInvalidResponseTier;
 }
@@ -4050,7 +4131,9 @@ var GEMINI_3X_MIN_THINKING_LEVEL = /* @__PURE__ */ new Map([
   ["gemini-3.6-flash", "minimal"],
   // L4 live evidence: Interactions gemini-3.7-flash accepts low/medium/high,
   // but rejects minimal. Keep this exact entry isolated from legacy aliases.
-  ["gemini-3.7-flash", "low"]
+  ["gemini-3.7-flash", "low"],
+  // Current-model live evidence: gemini-3.8-flash has the same low floor.
+  ["gemini-3.8-flash", "low"]
 ]);
 var GEMINI_MODELS_WITHOUT_PORTABLE_TEMPERATURE = /* @__PURE__ */ new Set([
   "gemini-3.5-flash-lite",
@@ -4078,10 +4161,11 @@ function warnGemini3xEffortNone(model) {
 var GeminiProvider = class {
   name = "google";
   inputCapabilities(model) {
-    return GEMINI_INTERACTIONS_IMAGE_MODELS.has(model) ? { image: { sources: ["bytes", "base64", "provider-file"] } } : {};
+    return model.trim().length > 0 ? { image: { sources: ["bytes", "base64", "provider-file"] } } : {};
   }
   validateInput(request) {
-    const effectiveModel = typeof request.providerOptions?.model === "string" ? request.providerOptions.model : request.model;
+    const modelOverride = request.providerOptions?.model;
+    const effectiveModel = typeof modelOverride === "string" ? modelOverride : request.model;
     const fail = (source, feature) => {
       throw new UnsupportedModelInputError({
         provider: this.name,
@@ -4091,7 +4175,10 @@ var GeminiProvider = class {
         ...feature ? { feature } : {}
       });
     };
-    if (!GEMINI_INTERACTIONS_IMAGE_MODELS.has(effectiveModel)) {
+    if (request.providerOptions && "model" in request.providerOptions && (typeof modelOverride !== "string" || modelOverride.trim().length === 0)) {
+      fail(void 0, "invalid model providerOptions");
+    }
+    if (effectiveModel.trim().length === 0) {
       fail(void 0, "image input for this model");
     }
     if (request.providerOptions) {
@@ -4283,7 +4370,7 @@ var GeminiProvider = class {
       });
     }
     if (!res.body) throw new Error("Gemini Interactions stream has no body");
-    yield* this.parseInteractionSSEStream(res.body);
+    yield* this.parseInteractionSSEStream(res.body, res.headers);
   }
   buildInteractionRequestBody(messages, options, stream) {
     const model = this.requestModel(options);
@@ -4353,6 +4440,7 @@ var GeminiProvider = class {
     return body;
   }
   assertSafeInteractionOptions(options) {
+    const modelOverride = options.providerOptions?.model;
     const model = this.requestModel(options);
     const fail = (feature) => {
       throw new UnsupportedModelInputError({
@@ -4362,7 +4450,10 @@ var GeminiProvider = class {
         feature
       });
     };
-    if (!GEMINI_INTERACTIONS_IMAGE_MODELS.has(model)) fail("image input for this model");
+    if (options.providerOptions && "model" in options.providerOptions && (typeof modelOverride !== "string" || modelOverride.trim().length === 0)) {
+      fail("invalid model providerOptions");
+    }
+    if (model.trim().length === 0) fail("image input for this model");
     const forbidden = [
       "input",
       "contents",
@@ -4438,7 +4529,7 @@ var GeminiProvider = class {
       providerMetadata: json.steps?.length ? { geminiInteractionSteps: json.steps.filter(isGeminiInteractionStep) } : void 0
     };
   }
-  async *parseInteractionSSEStream(body) {
+  async *parseInteractionSSEStream(body, responseHeaders) {
     const reader = body.getReader();
     const decoder = new TextDecoder();
     let buffer = "";
@@ -4464,7 +4555,22 @@ var GeminiProvider = class {
             continue;
           }
           if (event.event_type === "error") {
-            throw new Error(event.error?.message ?? "Gemini Interactions stream failed");
+            const status = geminiInteractionErrorStatus(event.error?.code);
+            const message = event.error?.message ?? "Gemini Interactions stream failed";
+            if (status !== void 0) {
+              throw buildProviderError({
+                provider: this.name,
+                status,
+                headers: responseHeaders,
+                message
+              });
+            }
+            throw new ProviderError({
+              provider: this.name,
+              status: 0,
+              retryable: false,
+              message
+            });
           }
           if (event.event_type === "step.start" && event.step) {
             const index = event.index;
@@ -5783,7 +5889,6 @@ function normalizeTranscriptionAccounting(usageValue, pricingValue) {
 // src/providers/transcription-utils.ts
 var OPENAI_TRANSCRIPTION_MODEL = "gpt-transcribe";
 var GEMINI_TRANSCRIPTION_MODEL = "gemini-3.5-transcribe";
-var OPENROUTER_TRANSCRIPTION_MODEL = "openai/whisper-1";
 function requireExactTranscriptionModel(provider, model, expected) {
   if (model !== expected) {
     throw new UnsupportedTranscriptionInputError({
@@ -6434,10 +6539,16 @@ var OpenRouterTranscriptionProvider = class {
     this.governor = options.rateLimit ? new RateLimiter(options.rateLimit) : void 0;
   }
   capabilities(model) {
-    return model === OPENROUTER_TRANSCRIPTION_MODEL ? { sources: ["bytes", "base64"] } : void 0;
+    return model.trim().length > 0 ? { sources: ["bytes", "base64"] } : void 0;
   }
   async transcribe(request) {
-    requireExactTranscriptionModel(this.name, request.model, OPENROUTER_TRANSCRIPTION_MODEL);
+    if (request.model.trim().length === 0) {
+      throw new UnsupportedTranscriptionInputError({
+        provider: this.name,
+        model: request.model,
+        feature: "this transcription model"
+      });
+    }
     requireNoProviderFile(request.audio, this.name, request.model);
     if (request.audio.type === "provider-file")
       throw new UnsupportedTranscriptionInputError({
@@ -6577,7 +6688,7 @@ var TranscriptionProviderRegistry = class {
   }
   resolve(uri, config = {}) {
     const colon = typeof uri === "string" ? uri.indexOf(":") : -1;
-    if (colon <= 0 || colon === uri.length - 1 || uri.indexOf(":", colon + 1) !== -1) {
+    if (colon <= 0 || colon === uri.length - 1) {
       throw new UnsupportedTranscriptionInputError({
         provider: "unknown",
         model: "",