@threaded/ai 1.0.25 → 1.0.27

package/dist/index.d.cts

@@ -93,6 +93,9 @@ type StreamEvent = {
     type: 'approval_requested';
     call: ToolCall;
     requestId: string;
+} | {
+    type: 'usage';
+    usage: TokenUsage;
 };
 interface ConversationContext {
     history: Message[];
@@ -108,6 +111,7 @@ interface ConversationContext {
     toolLimits?: Record<string, number>;
     toolConfig?: ToolExecutionConfig;
     abortSignal?: AbortSignal;
+    usage?: TokenUsage;
 }
 declare enum Inherit {
     Nothing = 0,
@@ -183,6 +187,11 @@ interface ImageResult {
     data: string;
     revisedPrompt?: string;
 }
+interface TokenUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+}
 
 declare const createMCPTools: (client: Client) => Promise<ToolConfig[]>;
 
@@ -191,6 +200,7 @@ declare const parseModelName: (model: string) => ParsedModel;
 declare const setKeys: (keys: ApiKeys) => void;
 declare const getKey: (provider: string) => string;
 declare const maxCalls: (toolConfig: ToolConfig, maxCalls: number) => ToolConfig;
+declare const addUsage: (existing: TokenUsage | undefined, promptTokens: number, completionTokens: number, totalTokens: number) => TokenUsage;
 
 /**
  * generates embeddings for text using openai or huggingface models
@@ -337,4 +347,4 @@ declare const rateLimited: (config: RateLimitConfig) => <T extends (...args: any
 
 declare const compose: (...steps: StepFunction[]) => ComposedFunction;
 
-export { type ApiKeys, type ApprovalRequest, type ApprovalResponse, type ComposedFunction, type ConfigOption, type ConversationContext, IMAGE_EDIT_MODEL_SCHEMA, IMAGE_MODEL_SCHEMA, type ImageConfig, type ImageModelSchema, type ImageResult, Inherit, type JsonSchema, type Message, type ModelConfig, type ParsedModel, type ProviderConfig, type ProviderModels, type RetryOptions, type SchemaProperty, type ScopeConfig, type StandardSchema, type StepFunction, type StreamEvent, type Thread, type ThreadStore, type ToolCall, type ToolCallResult, type ToolConfig, type ToolDefinition, type ToolExecutionConfig, appendToLastRequest, compose, convertMCPSchemaToToolSchema, convertStandardSchemaToJsonSchema, convertStandardSchemaToSchemaProperties, createMCPTools, embed, everyNMessages, everyNTokens, generateApprovalToken, generateImage, getDefaultConfig, getKey, getModelConfig, getOrCreateThread, isStandardSchema, maxCalls, model, noToolsCalled, normalizeSchema, onApprovalRequested, onApprovalResolved, parseModelName, rateLimited, removeApprovalListener, requestApproval, resolveApproval, retry, scope, setKeys, tap, toolConfigToToolDefinition, toolNotUsedInNTurns, toolWasCalled, when };
+export { type ApiKeys, type ApprovalRequest, type ApprovalResponse, type ComposedFunction, type ConfigOption, type ConversationContext, IMAGE_EDIT_MODEL_SCHEMA, IMAGE_MODEL_SCHEMA, type ImageConfig, type ImageModelSchema, type ImageResult, Inherit, type JsonSchema, type Message, type ModelConfig, type ParsedModel, type ProviderConfig, type ProviderModels, type RetryOptions, type SchemaProperty, type ScopeConfig, type StandardSchema, type StepFunction, type StreamEvent, type Thread, type ThreadStore, type TokenUsage, type ToolCall, type ToolCallResult, type ToolConfig, type ToolDefinition, type ToolExecutionConfig, addUsage, appendToLastRequest, compose, convertMCPSchemaToToolSchema, convertStandardSchemaToJsonSchema, convertStandardSchemaToSchemaProperties, createMCPTools, embed, everyNMessages, everyNTokens, generateApprovalToken, generateImage, getDefaultConfig, getKey, getModelConfig, getOrCreateThread, isStandardSchema, maxCalls, model, noToolsCalled, normalizeSchema, onApprovalRequested, onApprovalResolved, parseModelName, rateLimited, removeApprovalListener, requestApproval, resolveApproval, retry, scope, setKeys, tap, toolConfigToToolDefinition, toolNotUsedInNTurns, toolWasCalled, when };

package/dist/index.d.ts

@@ -93,6 +93,9 @@ type StreamEvent = {
     type: 'approval_requested';
     call: ToolCall;
     requestId: string;
+} | {
+    type: 'usage';
+    usage: TokenUsage;
 };
 interface ConversationContext {
     history: Message[];
@@ -108,6 +111,7 @@ interface ConversationContext {
     toolLimits?: Record<string, number>;
     toolConfig?: ToolExecutionConfig;
     abortSignal?: AbortSignal;
+    usage?: TokenUsage;
 }
 declare enum Inherit {
     Nothing = 0,
@@ -183,6 +187,11 @@ interface ImageResult {
     data: string;
     revisedPrompt?: string;
 }
+interface TokenUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+}
 
 declare const createMCPTools: (client: Client) => Promise<ToolConfig[]>;
 
@@ -191,6 +200,7 @@ declare const parseModelName: (model: string) => ParsedModel;
 declare const setKeys: (keys: ApiKeys) => void;
 declare const getKey: (provider: string) => string;
 declare const maxCalls: (toolConfig: ToolConfig, maxCalls: number) => ToolConfig;
+declare const addUsage: (existing: TokenUsage | undefined, promptTokens: number, completionTokens: number, totalTokens: number) => TokenUsage;
 
 /**
  * generates embeddings for text using openai or huggingface models
@@ -337,4 +347,4 @@ declare const rateLimited: (config: RateLimitConfig) => <T extends (...args: any
 
 declare const compose: (...steps: StepFunction[]) => ComposedFunction;
 
-export { type ApiKeys, type ApprovalRequest, type ApprovalResponse, type ComposedFunction, type ConfigOption, type ConversationContext, IMAGE_EDIT_MODEL_SCHEMA, IMAGE_MODEL_SCHEMA, type ImageConfig, type ImageModelSchema, type ImageResult, Inherit, type JsonSchema, type Message, type ModelConfig, type ParsedModel, type ProviderConfig, type ProviderModels, type RetryOptions, type SchemaProperty, type ScopeConfig, type StandardSchema, type StepFunction, type StreamEvent, type Thread, type ThreadStore, type ToolCall, type ToolCallResult, type ToolConfig, type ToolDefinition, type ToolExecutionConfig, appendToLastRequest, compose, convertMCPSchemaToToolSchema, convertStandardSchemaToJsonSchema, convertStandardSchemaToSchemaProperties, createMCPTools, embed, everyNMessages, everyNTokens, generateApprovalToken, generateImage, getDefaultConfig, getKey, getModelConfig, getOrCreateThread, isStandardSchema, maxCalls, model, noToolsCalled, normalizeSchema, onApprovalRequested, onApprovalResolved, parseModelName, rateLimited, removeApprovalListener, requestApproval, resolveApproval, retry, scope, setKeys, tap, toolConfigToToolDefinition, toolNotUsedInNTurns, toolWasCalled, when };
+export { type ApiKeys, type ApprovalRequest, type ApprovalResponse, type ComposedFunction, type ConfigOption, type ConversationContext, IMAGE_EDIT_MODEL_SCHEMA, IMAGE_MODEL_SCHEMA, type ImageConfig, type ImageModelSchema, type ImageResult, Inherit, type JsonSchema, type Message, type ModelConfig, type ParsedModel, type ProviderConfig, type ProviderModels, type RetryOptions, type SchemaProperty, type ScopeConfig, type StandardSchema, type StepFunction, type StreamEvent, type Thread, type ThreadStore, type TokenUsage, type ToolCall, type ToolCallResult, type ToolConfig, type ToolDefinition, type ToolExecutionConfig, addUsage, appendToLastRequest, compose, convertMCPSchemaToToolSchema, convertStandardSchemaToJsonSchema, convertStandardSchemaToSchemaProperties, createMCPTools, embed, everyNMessages, everyNTokens, generateApprovalToken, generateImage, getDefaultConfig, getKey, getModelConfig, getOrCreateThread, isStandardSchema, maxCalls, model, noToolsCalled, normalizeSchema, onApprovalRequested, onApprovalResolved, parseModelName, rateLimited, removeApprovalListener, requestApproval, resolveApproval, retry, scope, setKeys, tap, toolConfigToToolDefinition, toolNotUsedInNTurns, toolWasCalled, when };

package/dist/index.js

@@ -149,6 +149,11 @@ var maxCalls = (toolConfig, maxCalls2) => ({
   ...toolConfig,
   _maxCalls: maxCalls2
 });
+var addUsage = (existing, promptTokens, completionTokens, totalTokens) => ({
+  promptTokens: (existing?.promptTokens || 0) + promptTokens,
+  completionTokens: (existing?.completionTokens || 0) + completionTokens,
+  totalTokens: (existing?.totalTokens || 0) + totalTokens
+});
 
 // src/embed.ts
 var modelCache = /* @__PURE__ */ new Map();
@@ -460,7 +465,8 @@ var callOpenAI = async (config, ctx) => {
   const body = {
     model: model2,
     messages,
-    stream: !!ctx.stream
+    stream: !!ctx.stream,
+    ...ctx.stream && { stream_options: { include_usage: true } }
   };
   if (schema) {
     body.response_format = {
@@ -505,7 +511,8 @@ var callOpenAI = async (config, ctx) => {
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage: addUsage(ctx.usage, data.usage?.prompt_tokens || 0, data.usage?.completion_tokens || 0, data.usage?.total_tokens || 0)
   };
 };
 var handleOpenAIStream = async (response, ctx) => {
@@ -514,6 +521,7 @@ var handleOpenAIStream = async (response, ctx) => {
   let fullContent = "";
   let toolCalls = [];
   let buffer = "";
+  let streamUsage = null;
   try {
     while (true) {
       if (ctx.abortSignal?.aborted) {
@@ -531,6 +539,9 @@ var handleOpenAIStream = async (response, ctx) => {
           if (!data) continue;
           try {
             const parsed = JSON.parse(data);
+            if (parsed.usage) {
+              streamUsage = parsed.usage;
+            }
             const delta = parsed.choices?.[0]?.delta;
             if (delta?.content) {
               fullContent += delta.content;
@@ -556,10 +567,15 @@ var handleOpenAIStream = async (response, ctx) => {
   if (toolCalls.length > 0) {
     msg.tool_calls = toolCalls;
   }
+  const usage = addUsage(ctx.usage, streamUsage?.prompt_tokens || 0, streamUsage?.completion_tokens || 0, streamUsage?.total_tokens || 0);
+  if (ctx.stream && streamUsage) {
+    ctx.stream({ type: "usage", usage });
+  }
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage
   };
 };
 
@@ -695,10 +711,13 @@ Return only the JSON object, no other text or formatting.`;
       }
     ];
   }
+  const inputTokens = data.usage?.input_tokens || 0;
+  const outputTokens = data.usage?.output_tokens || 0;
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage: addUsage(ctx.usage, inputTokens, outputTokens, inputTokens + outputTokens)
   };
 };
 var handleAnthropicStream = async (response, ctx) => {
@@ -707,6 +726,8 @@ var handleAnthropicStream = async (response, ctx) => {
   let fullContent = "";
   const toolCalls = [];
   let buffer = "";
+  let inputTokens = 0;
+  let outputTokens = 0;
   try {
     while (true) {
       if (ctx.abortSignal?.aborted) {
@@ -723,6 +744,12 @@ var handleAnthropicStream = async (response, ctx) => {
           if (!data) continue;
           try {
             const parsed = JSON.parse(data);
+            if (parsed.type === "message_start" && parsed.message?.usage) {
+              inputTokens = parsed.message.usage.input_tokens || 0;
+            }
+            if (parsed.type === "message_delta" && parsed.usage) {
+              outputTokens = parsed.usage.output_tokens || 0;
+            }
             if (parsed.type === "content_block_delta" && parsed.delta?.text) {
               fullContent += parsed.delta.text;
               if (ctx.stream) {
@@ -762,10 +789,15 @@ var handleAnthropicStream = async (response, ctx) => {
   if (toolCalls.length > 0) {
     msg.tool_calls = toolCalls.map(({ index, ...tc }) => tc);
   }
+  const usage = addUsage(ctx.usage, inputTokens, outputTokens, inputTokens + outputTokens);
+  if (ctx.stream && (inputTokens || outputTokens)) {
+    ctx.stream({ type: "usage", usage });
+  }
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage
   };
 };
 
@@ -918,10 +950,12 @@ var callGoogle = async (config, ctx) => {
   if (toolCalls.length > 0) {
     msg.tool_calls = toolCalls;
   }
+  const um = data.usageMetadata;
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage: addUsage(ctx.usage, um?.promptTokenCount || 0, um?.candidatesTokenCount || 0, um?.totalTokenCount || 0)
   };
 };
 var handleGoogleStream = async (response, ctx) => {
@@ -930,6 +964,7 @@ var handleGoogleStream = async (response, ctx) => {
   let fullContent = "";
   const toolCalls = [];
   let buffer = "";
+  let usageMetadata = null;
   try {
     while (true) {
       if (ctx.abortSignal?.aborted) {
@@ -946,6 +981,9 @@ var handleGoogleStream = async (response, ctx) => {
           if (!data) continue;
           try {
             const parsed = JSON.parse(data);
+            if (parsed.usageMetadata) {
+              usageMetadata = parsed.usageMetadata;
+            }
             const candidate = parsed.candidates?.[0];
             const parts = candidate?.content?.parts || [];
             for (const part of parts) {
@@ -985,18 +1023,75 @@ var handleGoogleStream = async (response, ctx) => {
   if (toolCalls.length > 0) {
     msg.tool_calls = toolCalls;
   }
+  const um = usageMetadata;
+  const usage = addUsage(ctx.usage, um?.promptTokenCount || 0, um?.candidatesTokenCount || 0, um?.totalTokenCount || 0);
+  if (ctx.stream && um) {
+    ctx.stream({ type: "usage", usage });
+  }
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage
   };
 };
 
 // src/providers/huggingface.ts
+var modelCache2 = /* @__PURE__ */ new Map();
+var formatMessages = (instructions, history) => {
+  const messages = [];
+  if (instructions) {
+    messages.push({ role: "system", content: instructions });
+  }
+  for (const msg of history) {
+    messages.push({ role: msg.role, content: msg.content });
+  }
+  return messages;
+};
 var callHuggingFace = async (config, ctx) => {
-  throw new Error(
-    "Hugging Face provider not yet implemented. Use openai/, anthropic/, or google/ prefixes."
-  );
+  const { model: model2, instructions, schema } = config;
+  const { pipeline } = await import("@huggingface/transformers");
+  if (!modelCache2.has(model2)) {
+    const generator2 = await pipeline("text-generation", model2, {
+      dtype: "q4f16"
+    });
+    modelCache2.set(model2, generator2);
+  }
+  const generator = modelCache2.get(model2);
+  const messages = formatMessages(instructions, ctx.history);
+  if (schema) {
+    const schemaMsg = messages.find((m) => m.role === "system");
+    const schemaInstructions = [
+      "you must respond with valid JSON matching this schema:",
+      JSON.stringify(schema.schema, null, 2),
+      "respond ONLY with the JSON object, no other text."
+    ].join("\n");
+    if (schemaMsg) {
+      schemaMsg.content += "\n\n" + schemaInstructions;
+    } else {
+      messages.unshift({ role: "system", content: schemaInstructions });
+    }
+  }
+  const output = await generator(messages, {
+    max_new_tokens: 2048,
+    do_sample: false
+  });
+  const generatedMessages = output[0].generated_text;
+  const lastMessage = generatedMessages.at(-1);
+  const content = lastMessage?.content || "";
+  const msg = {
+    role: "assistant",
+    content
+  };
+  if (ctx.stream) {
+    ctx.stream({ type: "content", content });
+  }
+  return {
+    ...ctx,
+    lastResponse: msg,
+    history: [...ctx.history, msg],
+    usage: addUsage(ctx.usage, 0, 0, 0)
+  };
 };
 
 // src/providers/xai.ts
@@ -1037,7 +1132,8 @@ var callXAI = async (config, ctx) => {
   const body = {
     model: model2,
     messages,
-    stream: !!ctx.stream
+    stream: !!ctx.stream,
+    ...ctx.stream && { stream_options: { include_usage: true } }
   };
   if (schema) {
     body.response_format = {
@@ -1082,7 +1178,8 @@ var callXAI = async (config, ctx) => {
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage: addUsage(ctx.usage, data.usage?.prompt_tokens || 0, data.usage?.completion_tokens || 0, data.usage?.total_tokens || 0)
   };
 };
 var handleXAIStream = async (response, ctx) => {
@@ -1091,6 +1188,7 @@ var handleXAIStream = async (response, ctx) => {
   let fullContent = "";
   let toolCalls = [];
   let buffer = "";
+  let streamUsage = null;
   try {
     while (true) {
       if (ctx.abortSignal?.aborted) {
@@ -1108,6 +1206,9 @@ var handleXAIStream = async (response, ctx) => {
           if (!data) continue;
           try {
             const parsed = JSON.parse(data);
+            if (parsed.usage) {
+              streamUsage = parsed.usage;
+            }
             const delta = parsed.choices?.[0]?.delta;
             if (delta?.content) {
               fullContent += delta.content;
@@ -1133,10 +1234,15 @@ var handleXAIStream = async (response, ctx) => {
   if (toolCalls.length > 0) {
     msg.tool_calls = toolCalls;
   }
+  const usage = addUsage(ctx.usage, streamUsage?.prompt_tokens || 0, streamUsage?.completion_tokens || 0, streamUsage?.total_tokens || 0);
+  if (ctx.stream && streamUsage) {
+    ctx.stream({ type: "usage", usage });
+  }
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage
   };
 };
 
@@ -1169,7 +1275,8 @@ var callLocal = async (config, ctx) => {
   const body = {
     model: model2,
     messages,
-    stream: !!ctx.stream
+    stream: !!ctx.stream,
+    ...ctx.stream && { stream_options: { include_usage: true } }
   };
   if (schema) {
     body.response_format = {
@@ -1217,7 +1324,8 @@ var callLocal = async (config, ctx) => {
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage: addUsage(ctx.usage, data.usage?.prompt_tokens || 0, data.usage?.completion_tokens || 0, data.usage?.total_tokens || 0)
   };
 };
 var handleLocalStream = async (response, ctx) => {
@@ -1226,6 +1334,7 @@ var handleLocalStream = async (response, ctx) => {
   let fullContent = "";
   let toolCalls = [];
   let buffer = "";
+  let streamUsage = null;
   try {
     while (true) {
       if (ctx.abortSignal?.aborted) {
@@ -1243,6 +1352,9 @@ var handleLocalStream = async (response, ctx) => {
           if (!data) continue;
           try {
             const parsed = JSON.parse(data);
+            if (parsed.usage) {
+              streamUsage = parsed.usage;
+            }
             const delta = parsed.choices?.[0]?.delta;
             if (delta?.content) {
               fullContent += delta.content;
@@ -1268,10 +1380,15 @@ var handleLocalStream = async (response, ctx) => {
   if (toolCalls.length > 0) {
     msg.tool_calls = toolCalls;
   }
+  const usage = addUsage(ctx.usage, streamUsage?.prompt_tokens || 0, streamUsage?.completion_tokens || 0, streamUsage?.total_tokens || 0);
+  if (ctx.stream && streamUsage) {
+    ctx.stream({ type: "usage", usage });
+  }
   return {
     ...ctx,
     lastResponse: msg,
-    history: [...ctx.history, msg]
+    history: [...ctx.history, msg],
+    usage
   };
 };
 
@@ -1788,6 +1905,7 @@ var scopeContext = (config, ctx) => {
   }
   scopedCtx.stream = ctx.stream;
   scopedCtx.abortSignal = ctx.abortSignal;
+  scopedCtx.usage = ctx.usage;
   if (config.tools) {
     const toolDefinitions = config.tools.map(toolConfigToToolDefinition);
     const toolExecutors = config.tools.reduce(
@@ -1841,7 +1959,8 @@ var scope = (config, ...steps) => {
       history: config.silent ? ctx.history : scopedCtx.history,
       lastResponse: config.silent ? ctx.lastResponse : scopedCtx.lastResponse,
       lastRequest: config.silent ? ctx.lastRequest : scopedCtx.lastRequest,
-      stopReason: config.silent ? ctx.stopReason : scopedCtx.stopReason
+      stopReason: config.silent ? ctx.stopReason : scopedCtx.stopReason,
+      usage: scopedCtx.usage
     };
   };
 };
@@ -1902,6 +2021,7 @@ export {
   IMAGE_EDIT_MODEL_SCHEMA,
   IMAGE_MODEL_SCHEMA,
   Inherit,
+  addUsage,
   appendToLastRequest,
   compose,
   convertMCPSchemaToToolSchema,