@ljoukov/llm 7.0.10 → 7.0.12

package/README.md

@@ -450,19 +450,19 @@ Use a `chatgpt-` prefix:
 import { generateText } from "@ljoukov/llm";
 
 const result = await generateText({
-  model: "chatgpt-gpt-5.4",
+  model: "chatgpt-gpt-5.5",
   input: "Return exactly: OK",
 });
 
 console.log(result.text);
 ```
 
-`chatgpt-gpt-5.4-fast` is also supported as a convenience alias for ChatGPT-authenticated `gpt-5.4` with priority processing enabled (`service_tier="priority"`), matching Codex `/fast` semantics.
+`gpt-5.5-fast` and `chatgpt-gpt-5.5-fast` are also supported as convenience aliases for `gpt-5.5` with priority processing enabled (`service_tier="priority"`), matching Codex `/fast` semantics.
 
 Supported OpenAI text model ids are fixed literal unions in code, not arbitrary strings:
 
-- OpenAI API: `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`
-- ChatGPT auth: `chatgpt-gpt-5.4`, `chatgpt-gpt-5.4-fast`, `chatgpt-gpt-5.4-mini`, `chatgpt-gpt-5.3-codex-spark`
+- OpenAI API: `gpt-5.5`, `gpt-5.5-fast`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`
+- ChatGPT auth: `chatgpt-gpt-5.5`, `chatgpt-gpt-5.5-fast`, `chatgpt-gpt-5.4`, `chatgpt-gpt-5.4-fast`, `chatgpt-gpt-5.4-mini`, `chatgpt-gpt-5.3-codex-spark`
 
 ## JSON outputs
 
@@ -628,20 +628,22 @@ Architecture note:
 
 ### Provider-Native Tools (`generateText()`)
 
-Use this when the model provider executes the tool remotely (for example search/code-exec style tools).
+Use this when the model provider executes the tool remotely (for example search/code-exec/shell style tools).
 
 ```ts
 import { generateText } from "@ljoukov/llm";
 
 const result = await generateText({
-  model: "gpt-5.4-mini",
-  input: "Find 3 relevant sources about X and summarize them.",
-  tools: [{ type: "web-search", mode: "live" }, { type: "code-execution" }],
+  model: "gpt-5.5",
+  input: "Run python --version in a hosted shell and summarize the result.",
+  tools: [{ type: "shell" }],
 });
 
 console.log(result.text);
 ```
 
+`{ type: "shell" }` uses OpenAI hosted shell containers by default. It is only supported for OpenAI API models; ChatGPT-authenticated, Gemini, and Fireworks providers reject it.
+
 ### Runtime Tools (`runToolLoop()`)
 
 Use this when the model should call your local runtime functions.
@@ -742,7 +744,7 @@ console.log(result.text);
 For read/search/write tasks in a workspace, enable `filesystemTool`. The library auto-selects a tool profile by model
 when `profile: "auto"`:
 
-- Codex-like models (`gpt-5.4`, `chatgpt-gpt-5.4`, `chatgpt-gpt-5.4-fast`, and `chatgpt-gpt-5.3-codex-spark`): Codex-compatible filesystem tool shape.
+- Codex-like models (`gpt-5.5`, `gpt-5.5-fast`, `chatgpt-gpt-5.5`, `chatgpt-gpt-5.5-fast`, `gpt-5.4`, `chatgpt-gpt-5.4`, `chatgpt-gpt-5.4-fast`, and `chatgpt-gpt-5.3-codex-spark`): Codex-compatible filesystem tool shape.
 - Gemini models: Gemini-compatible filesystem tool shape.
 - Other models: model-agnostic profile (currently Gemini-style).
 

package/dist/index.cjs

@@ -321,16 +321,27 @@ function getGeminiImagePricing(modelId) {
 }
 
 // src/openai/models.ts
-var OPENAI_MODEL_IDS = ["gpt-5.4", "gpt-5.4-mini", "gpt-5.4-nano"];
+var OPENAI_MODEL_IDS = [
+  "gpt-5.5",
+  "gpt-5.5-fast",
+  "gpt-5.4",
+  "gpt-5.4-mini",
+  "gpt-5.4-nano"
+];
 function isOpenAiModelId(value) {
   return OPENAI_MODEL_IDS.includes(value);
 }
 var CHATGPT_MODEL_IDS = [
+  "chatgpt-gpt-5.5",
+  "chatgpt-gpt-5.5-fast",
   "chatgpt-gpt-5.4",
   "chatgpt-gpt-5.4-fast",
   "chatgpt-gpt-5.4-mini",
   "chatgpt-gpt-5.3-codex-spark"
 ];
+var FAST_MODEL_SUFFIX = "-fast";
+var OPENAI_PRIORITY_MODEL_IDS = ["gpt-5.5-fast"];
+var CHATGPT_PRIORITY_MODEL_IDS = ["chatgpt-gpt-5.5-fast", "chatgpt-gpt-5.4-fast"];
 var EXPERIMENTAL_CHATGPT_MODEL_PREFIX = "experimental-chatgpt-";
 function isExperimentalChatGptModelId(value) {
   return value.startsWith(EXPERIMENTAL_CHATGPT_MODEL_PREFIX) && value.length > EXPERIMENTAL_CHATGPT_MODEL_PREFIX.length;
@@ -338,6 +349,15 @@ function isExperimentalChatGptModelId(value) {
 function isChatGptModelId(value) {
   return CHATGPT_MODEL_IDS.includes(value) || isExperimentalChatGptModelId(value);
 }
+function stripFastSuffix(model) {
+  return model.endsWith(FAST_MODEL_SUFFIX) ? model.slice(0, -FAST_MODEL_SUFFIX.length) : model;
+}
+function resolveOpenAiProviderModel(model) {
+  return OPENAI_PRIORITY_MODEL_IDS.includes(model) ? stripFastSuffix(model) : model;
+}
+function resolveOpenAiServiceTier(model) {
+  return OPENAI_PRIORITY_MODEL_IDS.includes(model) ? "priority" : void 0;
+}
 function stripChatGptPrefix(model) {
   if (isExperimentalChatGptModelId(model)) {
     return model.slice(EXPERIMENTAL_CHATGPT_MODEL_PREFIX.length);
@@ -345,18 +365,16 @@ function stripChatGptPrefix(model) {
   return model.slice("chatgpt-".length);
 }
 function resolveChatGptProviderModel(model) {
-  switch (model) {
-    case "chatgpt-gpt-5.4-fast":
-      return "gpt-5.4";
-    default:
-      return stripChatGptPrefix(model);
-  }
+  const providerModel = stripChatGptPrefix(model);
+  return CHATGPT_PRIORITY_MODEL_IDS.includes(model) ? stripFastSuffix(providerModel) : providerModel;
 }
 function resolveChatGptServiceTier(model) {
-  return model === "chatgpt-gpt-5.4-fast" ? "priority" : void 0;
+  return CHATGPT_PRIORITY_MODEL_IDS.includes(model) ? "priority" : void 0;
 }
 
 // src/openai/pricing.ts
+var OPENAI_GPT_55_FAST_MODEL_IDS = ["gpt-5.5-fast", "chatgpt-gpt-5.5-fast"];
+var OPENAI_GPT_55_STANDARD_MODEL_IDS = ["gpt-5.5", "chatgpt-gpt-5.5"];
 var OPENAI_GPT_54_FAST_MODEL_IDS = ["gpt-5.4-fast", "chatgpt-gpt-5.4-fast"];
 var OPENAI_GPT_54_MINI_MODEL_IDS = ["gpt-5.4-mini", "chatgpt-gpt-5.4-mini"];
 var OPENAI_GPT_54_NANO_MODEL_IDS = ["gpt-5.4-nano"];
@@ -365,6 +383,16 @@ var OPENAI_GPT_53_CODEX_SPARK_MODEL_IDS = [
   "chatgpt-gpt-5.3-codex-spark"
 ];
 var OPENAI_GPT_54_STANDARD_MODEL_IDS = ["gpt-5.4", "chatgpt-gpt-5.4"];
+var OPENAI_GPT_55_PRICING = {
+  inputRate: 5 / 1e6,
+  cachedRate: 0.5 / 1e6,
+  outputRate: 30 / 1e6
+};
+var OPENAI_GPT_55_PRIORITY_PRICING = {
+  inputRate: 12.5 / 1e6,
+  cachedRate: 1.25 / 1e6,
+  outputRate: 75 / 1e6
+};
 var OPENAI_GPT_54_PRICING = {
   inputRate: 2.5 / 1e6,
   cachedRate: 0.25 / 1e6,
@@ -389,6 +417,12 @@ function getOpenAiPricing(modelId) {
   if (isExperimentalChatGptModelId(modelId)) {
     return OPENAI_GPT_54_PRICING;
   }
+  if (OPENAI_GPT_55_FAST_MODEL_IDS.includes(modelId)) {
+    return OPENAI_GPT_55_PRIORITY_PRICING;
+  }
+  if (OPENAI_GPT_55_STANDARD_MODEL_IDS.includes(modelId)) {
+    return OPENAI_GPT_55_PRICING;
+  }
   if (OPENAI_GPT_54_FAST_MODEL_IDS.includes(modelId)) {
     return OPENAI_GPT_54_PRIORITY_PRICING;
   }
@@ -5070,7 +5104,11 @@ function resolveProvider(model) {
     }
   }
   if (isOpenAiModelId(model)) {
-    return { provider: "openai", model };
+    return {
+      provider: "openai",
+      model: resolveOpenAiProviderModel(model),
+      serviceTier: resolveOpenAiServiceTier(model)
+    };
   }
   throw new Error(`Unsupported text model: ${model}`);
 }
@@ -6159,12 +6197,40 @@ function toGeminiTools(tools) {
         return { googleSearch: {} };
       case "code-execution":
         return { codeExecution: {} };
+      case "shell":
+        throw new Error("Gemini provider does not support the OpenAI shell tool.");
       default:
         throw new Error("Unsupported tool configuration");
     }
   });
 }
-function toOpenAiTools(tools) {
+function toOpenAiShellEnvironment(environment) {
+  if (environment?.type === "container-reference") {
+    return {
+      type: "container_reference",
+      container_id: environment.containerId
+    };
+  }
+  return {
+    type: "container_auto",
+    ...environment?.fileIds ? { file_ids: Array.from(environment.fileIds) } : {},
+    ...environment?.memoryLimit !== void 0 ? { memory_limit: environment.memoryLimit } : {},
+    ...environment?.networkPolicy ? {
+      network_policy: environment.networkPolicy.type === "allowlist" ? {
+        type: "allowlist",
+        allowed_domains: Array.from(environment.networkPolicy.allowedDomains),
+        ...environment.networkPolicy.domainSecrets ? {
+          domain_secrets: environment.networkPolicy.domainSecrets.map((secret) => ({
+            domain: secret.domain,
+            name: secret.name,
+            value: secret.value
+          }))
+        } : {}
+      } : { type: "disabled" }
+    } : {}
+  };
+}
+function toOpenAiTools(tools, options) {
   if (!tools || tools.length === 0) {
     return void 0;
   }
@@ -6177,6 +6243,15 @@ function toOpenAiTools(tools) {
       case "code-execution": {
         return { type: "code_interpreter", container: { type: "auto" } };
       }
+      case "shell": {
+        if (options.provider !== "openai") {
+          throw new Error("OpenAI shell tool is only supported for OpenAI API models.");
+        }
+        return {
+          type: "shell",
+          environment: toOpenAiShellEnvironment(tool2.environment)
+        };
+      }
       default:
         throw new Error("Unsupported tool configuration");
     }
@@ -7775,6 +7850,8 @@ async function runTextCall(params) {
   let responseRole;
   let latestUsage;
   let responseImages = 0;
+  let sawResponseDelta = false;
+  let sawThoughtDelta = false;
   const pushEvent = (event) => {
     queue.push(event);
     params.onEvent?.(event);
@@ -7785,8 +7862,10 @@ async function runTextCall(params) {
     }
     responseParts.push({ type: "text", text, ...channel === "thought" ? { thought: true } : {} });
     if (channel === "thought") {
+      sawThoughtDelta = true;
       callLogger?.appendThoughtDelta(text);
     } else {
+      sawResponseDelta = true;
       callLogger?.appendResponseDelta(text);
     }
     pushEvent({ type: "delta", channel, text });
@@ -7826,7 +7905,7 @@ async function runTextCall(params) {
           }),
           { model: request.model, provider: "openai" }
         );
-        const openAiTools = toOpenAiTools(request.tools);
+        const openAiTools = toOpenAiTools(request.tools, { provider: "openai" });
         const reasoningEffort = resolveOpenAiReasoningEffort(
           modelForProvider,
           request.thinkingLevel
@@ -7844,6 +7923,7 @@ async function runTextCall(params) {
             {
               model: modelForProvider,
               input: openAiInput,
+              ...providerInfo.serviceTier ? { service_tier: providerInfo.serviceTier } : {},
               reasoning,
               text: openAiTextConfig,
               ...openAiTools ? { tools: openAiTools } : {},
@@ -7886,12 +7966,17 @@ async function runTextCall(params) {
             );
           }
           latestUsage = extractOpenAiUsageTokens(finalResponse.usage);
-          if (responseParts.length === 0) {
+          if (!sawResponseDelta || !sawThoughtDelta) {
+            const needsResponseFallback = !sawResponseDelta;
+            const needsThoughtFallback = !sawThoughtDelta;
             const fallback = extractOpenAiResponseParts(finalResponse);
             blocked = blocked || fallback.blocked;
             for (const part of fallback.parts) {
               if (part.type === "text") {
-                pushDelta(part.thought === true ? "thought" : "response", part.text);
+                const channel = part.thought === true ? "thought" : "response";
+                if (channel === "response" && needsResponseFallback || channel === "thought" && needsThoughtFallback) {
+                  pushDelta(channel, part.text);
+                }
               } else if (part.type === "inlineData") {
                 pushInline(part.data, part.mimeType);
               }
@@ -7908,7 +7993,7 @@ async function runTextCall(params) {
           provider: "chatgpt"
         });
         const reasoningEffort = resolveOpenAiReasoningEffort(request.model, request.thinkingLevel);
-        const openAiTools = toOpenAiTools(request.tools);
+        const openAiTools = toOpenAiTools(request.tools, { provider: "chatgpt" });
         const requestPayload = {
           model: modelForProvider,
           store: false,
@@ -7927,18 +8012,18 @@ async function runTextCall(params) {
           },
           ...openAiTools ? { tools: openAiTools } : {}
         };
-        let sawResponseDelta = false;
-        let sawThoughtDelta = false;
+        let sawResponseDelta2 = false;
+        let sawThoughtDelta2 = false;
         const result2 = await collectChatGptCodexResponseWithRetry({
           request: requestPayload,
           signal,
           onDelta: (delta) => {
             if (delta.thoughtDelta) {
-              sawThoughtDelta = true;
+              sawThoughtDelta2 = true;
               pushDelta("thought", delta.thoughtDelta);
             }
             if (delta.textDelta) {
-              sawResponseDelta = true;
+              sawResponseDelta2 = true;
               pushDelta("response", delta.textDelta);
             }
           }
@@ -7954,10 +8039,10 @@ async function runTextCall(params) {
         latestUsage = extractChatGptUsageTokens(result2.usage);
         const fallbackText = typeof result2.text === "string" ? result2.text : "";
         const fallbackThoughts = typeof result2.reasoningSummaryText === "string" && result2.reasoningSummaryText.length > 0 ? result2.reasoningSummaryText : typeof result2.reasoningText === "string" ? result2.reasoningText : "";
-        if (!sawThoughtDelta && fallbackThoughts.length > 0) {
+        if (!sawThoughtDelta2 && fallbackThoughts.length > 0) {
           pushDelta("thought", fallbackThoughts);
         }
-        if (!sawResponseDelta && fallbackText.length > 0) {
+        if (!sawResponseDelta2 && fallbackText.length > 0) {
           pushDelta("response", fallbackText);
         }
       } else if (provider === "fireworks") {
@@ -8677,7 +8762,7 @@ async function runToolLoop(request) {
   try {
     if (providerInfo.provider === "openai") {
       const openAiAgentTools = buildOpenAiToolsFromToolSet(request.tools);
-      const openAiNativeTools = toOpenAiTools(request.modelTools);
+      const openAiNativeTools = toOpenAiTools(request.modelTools, { provider: "openai" });
       const openAiTools = openAiNativeTools ? [...openAiNativeTools, ...openAiAgentTools] : [...openAiAgentTools];
       const reasoningEffort = resolveOpenAiReasoningEffort(
         providerInfo.model,
@@ -8729,6 +8814,7 @@ async function runToolLoop(request) {
         const stepRequestPayload = {
           model: providerInfo.model,
           input: preparedInput,
+          ...providerInfo.serviceTier ? { service_tier: providerInfo.serviceTier } : {},
           ...previousResponseId ? { previous_response_id: previousResponseId } : {},
           ...openAiTools.length > 0 ? { tools: openAiTools } : {},
           ...openAiTools.length > 0 ? { parallel_tool_calls: true } : {},
@@ -8757,6 +8843,7 @@ async function runToolLoop(request) {
                 {
                   model: providerInfo.model,
                   input: preparedInput,
+                  ...providerInfo.serviceTier ? { service_tier: providerInfo.serviceTier } : {},
                   ...previousResponseId ? { previous_response_id: previousResponseId } : {},
                   ...openAiTools.length > 0 ? { tools: openAiTools } : {},
                   ...openAiTools.length > 0 ? { parallel_tool_calls: true } : {},
@@ -9078,7 +9165,7 @@ async function runToolLoop(request) {
     }
     if (providerInfo.provider === "chatgpt") {
       const openAiAgentTools = buildOpenAiToolsFromToolSet(request.tools);
-      const openAiNativeTools = toOpenAiTools(request.modelTools);
+      const openAiNativeTools = toOpenAiTools(request.modelTools, { provider: "chatgpt" });
       const openAiTools = openAiNativeTools ? [...openAiNativeTools, ...openAiAgentTools] : [...openAiAgentTools];
       const reasoningEffort = resolveOpenAiReasoningEffort(request.model, request.thinkingLevel);
       const toolLoopInput = toChatGptInput(contents, {
@@ -12768,7 +12855,7 @@ function isCodexModel(model) {
     return true;
   }
   const normalized = model.startsWith("chatgpt-") ? model.slice("chatgpt-".length) : model;
-  return normalized.includes("codex") || normalized === "gpt-5.4" || normalized === "gpt-5.4-fast";
+  return normalized.includes("codex") || normalized === "gpt-5.5" || normalized === "gpt-5.5-fast" || normalized === "gpt-5.4" || normalized === "gpt-5.4-fast";
 }
 function isGeminiModel(model) {
   return model.startsWith("gemini-");