@stacksjs/ai 0.75.85 → 0.75.87

package/dist/drivers/anthropic/index.d.ts

@@ -1,4 +1,4 @@
-import type { AIDriver, AIDriverConfig, AIMessage, AIResult, ChatCompletionOptions } from '../../types';
+import type { AIDriver, AIDriverConfig, AIMessage, AIResult, AIStreamEvent, ChatCompletionOptions } from '../../types';
 /**
  * Configure Anthropic globally
  */
@@ -10,9 +10,22 @@ export declare function createAnthropicDriver(config: AnthropicDriverConfig): AI
  */
 export declare function chat(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): Promise<AIResult>;
 /**
- * Stream chat completion
+ * A streamed chat completion, as events: text as it is written, each tool
+ * call once its arguments have streamed, then the whole result.
+ *
+ * A `tool_use` block streams its input as `input_json_delta` fragments, so a
+ * call is reported at its `content_block_stop`, whole. The structured-output
+ * tool is internal, as in `chat()`: its JSON streams as text. An `error`
+ * event (an overload mid-stream) throws rather than ending the stream as if
+ * the model had finished.
  */
-export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): AsyncGenerator<string>;
+export declare function streamChatEvents(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): AsyncGenerator<AIStreamEvent>;
+/**
+ * Stream chat completion: the text as it is written. Tool calls, usage and
+ * the stop reason are in the result the generator returns; iterate
+ * {@link streamChatEvents} to act on a tool call as soon as it arrives.
+ */
+export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): AsyncGenerator<string, AIResult, undefined>;
 /**
  * Simple prompt helper
  */
@@ -27,6 +40,7 @@ export declare const anthropic: {
   configure: typeof configure;
   chat: typeof chat;
   streamChat: typeof streamChat;
+  streamChatEvents: typeof streamChatEvents;
   prompt: typeof prompt;
   estimateTokens: typeof estimateTokens;
   createDriver: typeof createAnthropicDriver
@@ -37,4 +51,13 @@ export declare interface AnthropicDriverConfig extends AIDriverConfig {
   maxTokens?: number
   anthropicVersion?: string
 }
+/** The Messages API's streamed events, as far as a chat completion reads them. */
+declare type MessagesStreamEvent = | { type: 'message_start', message: { model: string, usage?: { input_tokens?: number, output_tokens?: number } } }
+  | { type: 'content_block_start', index: number, content_block: { type: string, id?: string, name?: string } }
+  | { type: 'content_block_delta', index: number, delta: { type: 'text_delta', text: string } | { type: 'input_json_delta', partial_json: string } | { type: string } }
+  | { type: 'content_block_stop', index: number }
+  | { type: 'message_delta', delta: { stop_reason?: string }, usage?: { output_tokens?: number } }
+  | { type: 'message_stop' }
+  | { type: 'ping' }
+  | { type: 'error', error: { type: string, message: string } }
 export default anthropic;

package/dist/drivers/anthropic/index.js

@@ -1,3 +1 @@
-import{fetchWithRetry}from"../../utils/retry";import{recordUsage}from"../../utils/usage";import{normalizeMessagesForProvider}from"../../utils/vision";import{DEFAULT_ANTHROPIC_MODEL}from"../../models";const DEFAULT_MODEL=DEFAULT_ANTHROPIC_MODEL,DEFAULT_MAX_TOKENS=4096,DEFAULT_VERSION="2023-06-01",BASE_URL="https://api.anthropic.com/v1";let globalConfig=null;export function configure(config){globalConfig=config}function getConfig(config){const merged={...globalConfig,...config};if(!merged.apiKey)merged.apiKey=process.env.ANTHROPIC_API_KEY||"";return merged}export function createAnthropicDriver(config){const{apiKey,model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,anthropicVersion=DEFAULT_VERSION}=config;return{name:"Claude API",async process(command,systemPrompt,history){if(!apiKey)throw Error("Anthropic API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":apiKey,"anthropic-version":anthropicVersion},body:JSON.stringify({model,max_tokens:maxTokens,system:systemPrompt,messages:[...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}const data=await response.json();if(!data.content||data.content.length===0)throw Error("Claude API returned empty content");return data.content[0].text},async*stream(command,systemPrompt,history){if(!apiKey)throw Error("Anthropic API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":apiKey,"anthropic-version":anthropicVersion},body:JSON.stringify({model,max_tokens:maxTokens,system:systemPrompt,stream:!0,messages:[...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}const reader=response.body?.getReader();if(!reader)throw Error("No response body");const decoder=new TextDecoder;let buffer="";const handlePayload=function*(data){if(data==="[DONE]")return;let event;try{event=JSON.parse(data)}catch{return}if(event.type==="error"){const msg=event.error?.message??JSON.stringify(event.error??event);throw Error(`[anthropic/stream] mid-stream error: ${msg}`)}if(event.type==="content_block_delta"&&event.delta?.text)yield event.delta.text};while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines)if(line.startsWith("data: "))yield*handlePayload(line.slice(6))}if(buffer.startsWith("data: "))yield*handlePayload(buffer.slice(6))}}}export async function chat(messages,options={}){const config=getConfig(),{model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,temperature,topP,stop,system,tools,toolChoice,responseFormat}=options,normalizedMessages=normalizeMessagesForProvider(messages,"anthropic"),startedAt=Date.now(),body={model,max_tokens:maxTokens,temperature,top_p:topP,stop_sequences:stop?Array.isArray(stop)?stop:[stop]:void 0,system,messages:normalizedMessages};if(tools&&tools.length>0){body.tools=tools.map((t)=>({name:t.name,description:t.description,input_schema:t.parameters??{type:"object",properties:{}}}));if(toolChoice!==void 0)body.tool_choice=mapAnthropicToolChoice(toolChoice)}if(responseFormat&&responseFormat.type!=="text"){const outputTool=buildAnthropicJsonTool(responseFormat),existing=Array.isArray(body.tools)?body.tools:[];body.tools=[...existing,outputTool];if(toolChoice===void 0)body.tool_choice={type:"tool",name:outputTool.name}}const response=await fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":config.apiKey,"anthropic-version":config.anthropicVersion||DEFAULT_VERSION},body:JSON.stringify(body)});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}const data=await response.json();if(!data.content||data.content.length===0)throw Error("Claude API returned empty content");const outputTool=responseFormat&&responseFormat.type!=="text"?buildAnthropicJsonTool(responseFormat).name:void 0,blocks=data.content,structured=outputTool?blocks.find((b)=>b.type==="tool_use"&&b.name===outputTool):void 0,toolCalls=blocks.filter((b)=>b.type==="tool_use"&&b!==structured).map((b)=>({id:b.id??"",name:b.name??"",arguments:b.input??{}})),result={content:structured?JSON.stringify(structured.input):blocks.filter((b)=>b.type==="text").map((b)=>b.text??"").join(""),...toolCalls.length>0?{toolCalls}:{},model:data.model,usage:{promptTokens:data.usage?.input_tokens||0,completionTokens:data.usage?.output_tokens||0,totalTokens:(data.usage?.input_tokens||0)+(data.usage?.output_tokens||0)},finishReason:data.stop_reason};recordUsage({provider:"anthropic",model:data.model,promptTokens:result.usage.promptTokens,completionTokens:result.usage.completionTokens,totalTokens:result.usage.totalTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()});return result}function mapAnthropicToolChoice(choice){if(choice==="auto")return{type:"auto"};if(choice==="required")return{type:"any"};if(choice==="none")return{type:"none"};return{type:"tool",name:choice.name}}function buildAnthropicJsonTool(format){if(format.type==="json_schema")return{name:format.json_schema.name,description:`Returns the result as JSON matching the '${format.json_schema.name}' schema.`,input_schema:format.json_schema.schema};return{name:"structured_output",description:"Returns the result as a JSON object.",input_schema:{type:"object",additionalProperties:!0}}}export async function*streamChat(messages,options={}){const config=getConfig(),{model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,temperature,topP,stop,system}=options,response=await fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":config.apiKey,"anthropic-version":config.anthropicVersion||DEFAULT_VERSION},body:JSON.stringify({model,max_tokens:maxTokens,temperature,top_p:topP,stop_sequences:stop?Array.isArray(stop)?stop:[stop]:void 0,system,stream:!0,messages:normalizeMessagesForProvider(messages,"anthropic")})});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}const reader=response.body?.getReader();if(!reader)throw Error("No response body");const decoder=new TextDecoder;let buffer="";while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines)if(line.startsWith("data: ")){const data=line.slice(6);if(data==="[DONE]")continue;try{const event=JSON.parse(data);if(event.type==="content_block_delta"&&event.delta?.text)yield event.delta.text}catch{}}}}export async function prompt(text,options={}){return(await chat([{role:"user",content:text}],options)).content}export function estimateTokens(text){return Math.ceil(text.length/4)}export const anthropicDriver={create:createAnthropicDriver},anthropic={configure,chat,streamChat,prompt,estimateTokens,createDriver:createAnthropicDriver};export default anthropic;
+import{fetchWithRetry}from"../../utils/retry";import{sseJson,streamText}from"../../utils/stream";import{parseToolArguments}from"../../utils/tools";import{recordUsage}from"../../utils/usage";import{normalizeMessagesForProvider}from"../../utils/vision";import{DEFAULT_ANTHROPIC_MODEL}from"../../models";const DEFAULT_MODEL=DEFAULT_ANTHROPIC_MODEL,DEFAULT_MAX_TOKENS=4096,DEFAULT_VERSION="2023-06-01",BASE_URL="https://api.anthropic.com/v1";let globalConfig=null;export function configure(config){globalConfig=config}function getConfig(config){const merged={...globalConfig,...config};if(!merged.apiKey)merged.apiKey=process.env.ANTHROPIC_API_KEY||"";return merged}export function createAnthropicDriver(config){const{apiKey,model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,anthropicVersion=DEFAULT_VERSION}=config;return{name:"Claude API",async process(command,systemPrompt,history){if(!apiKey)throw Error("Anthropic API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":apiKey,"anthropic-version":anthropicVersion},body:JSON.stringify({model,max_tokens:maxTokens,system:systemPrompt,messages:[...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}const data=await response.json();if(!data.content||data.content.length===0)throw Error("Claude API returned empty content");return data.content[0].text},async*stream(command,systemPrompt,history){if(!apiKey)throw Error("Anthropic API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":apiKey,"anthropic-version":anthropicVersion},body:JSON.stringify({model,max_tokens:maxTokens,system:systemPrompt,stream:!0,messages:[...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}if(!response.body)throw Error("No response body");for await(const event of sseJson(response.body,"Claude API")){if(event.type==="error"){const msg=event.error?.message??JSON.stringify(event.error??event);throw Error(`[anthropic/stream] mid-stream error: ${msg}`)}if(event.type==="content_block_delta"&&event.delta?.text)yield event.delta.text}}}}function messagesRequest(messages,options){const{model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,temperature,topP,stop,system,tools,toolChoice,responseFormat}=options,body={model,max_tokens:maxTokens,temperature,top_p:topP,stop_sequences:stop?Array.isArray(stop)?stop:[stop]:void 0,system,messages:normalizeMessagesForProvider(messages,"anthropic")};if(tools&&tools.length>0){body.tools=tools.map((t)=>({name:t.name,description:t.description,input_schema:t.parameters??{type:"object",properties:{}}}));if(toolChoice!==void 0)body.tool_choice=mapAnthropicToolChoice(toolChoice)}if(responseFormat&&responseFormat.type!=="text"){const outputTool=buildAnthropicJsonTool(responseFormat),existing=Array.isArray(body.tools)?body.tools:[];body.tools=[...existing,outputTool];if(toolChoice===void 0)body.tool_choice={type:"tool",name:outputTool.name};return{body,outputTool:outputTool.name}}return{body}}function postMessages(body){const config=getConfig();return fetchWithRetry(`${BASE_URL}/messages`,{method:"POST",headers:{"Content-Type":"application/json","x-api-key":config.apiKey,"anthropic-version":config.anthropicVersion||DEFAULT_VERSION},body:JSON.stringify(body)})}export async function chat(messages,options={}){const startedAt=Date.now(),{body,outputTool}=messagesRequest(messages,options),response=await postMessages(body);if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}const data=await response.json();if(!data.content||data.content.length===0)throw Error("Claude API returned empty content");const blocks=data.content,structured=outputTool?blocks.find((b)=>b.type==="tool_use"&&b.name===outputTool):void 0,toolCalls=blocks.filter((b)=>b.type==="tool_use"&&b!==structured).map((b)=>({id:b.id??"",name:b.name??"",arguments:b.input??{}})),result={content:structured?JSON.stringify(structured.input):blocks.filter((b)=>b.type==="text").map((b)=>b.text??"").join(""),...toolCalls.length>0?{toolCalls}:{},model:data.model,usage:{promptTokens:data.usage?.input_tokens||0,completionTokens:data.usage?.output_tokens||0,totalTokens:(data.usage?.input_tokens||0)+(data.usage?.output_tokens||0)},finishReason:data.stop_reason};recordUsage({provider:"anthropic",model:data.model,promptTokens:result.usage.promptTokens,completionTokens:result.usage.completionTokens,totalTokens:result.usage.totalTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()});return result}function mapAnthropicToolChoice(choice){if(choice==="auto")return{type:"auto"};if(choice==="required")return{type:"any"};if(choice==="none")return{type:"none"};return{type:"tool",name:choice.name}}function buildAnthropicJsonTool(format){if(format.type==="json_schema")return{name:format.json_schema.name,description:`Returns the result as JSON matching the '${format.json_schema.name}' schema.`,input_schema:format.json_schema.schema};return{name:"structured_output",description:"Returns the result as a JSON object.",input_schema:{type:"object",additionalProperties:!0}}}export async function*streamChatEvents(messages,options={}){const startedAt=Date.now(),{body,outputTool}=messagesRequest(messages,options),response=await postMessages({...body,stream:!0});if(!response.ok){const error=await response.text();throw Error(`Claude API error: ${error}`)}if(!response.body)throw Error("No response body");const toolBlocks=new Map,toolCalls=[];let content="",model=String(body.model),promptTokens=0,completionTokens=0,finishReason;for await(const event of sseJson(response.body,"Claude API"))switch(event.type){case"message_start":model=event.message.model||model;promptTokens=event.message.usage?.input_tokens??0;completionTokens=event.message.usage?.output_tokens??0;break;case"content_block_start":if(event.content_block.type==="tool_use"){const name=event.content_block.name??"";toolBlocks.set(event.index,{id:event.content_block.id??"",name,json:"",structured:name===outputTool})}break;case"content_block_delta":{const delta=event.delta;if("text"in delta&&delta.type==="text_delta"){content+=delta.text;yield{type:"text",text:delta.text}}else if("partial_json"in delta&&delta.type==="input_json_delta"){const block=toolBlocks.get(event.index);if(!block)break;block.json+=delta.partial_json;if(block.structured&&delta.partial_json){content+=delta.partial_json;yield{type:"text",text:delta.partial_json}}}break}case"content_block_stop":{const block=toolBlocks.get(event.index);if(block&&!block.structured){const call={id:block.id,name:block.name,arguments:parseToolArguments(block.json,block.name)};toolCalls.push(call);yield{type:"tool_call",call}}break}case"message_delta":finishReason=event.delta.stop_reason??finishReason;completionTokens=event.usage?.output_tokens??completionTokens;break;case"error":throw Error(`Claude API error: ${event.error.type}: ${event.error.message}`)}const result={content,...toolCalls.length>0?{toolCalls}:{},model,usage:{promptTokens,completionTokens,totalTokens:promptTokens+completionTokens},...finishReason?{finishReason}:{}};recordUsage({provider:"anthropic",model,promptTokens,completionTokens,totalTokens:promptTokens+completionTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()});yield{type:"done",result}}export function streamChat(messages,options={}){return streamText(streamChatEvents(messages,options))}export async function prompt(text,options={}){return(await chat([{role:"user",content:text}],options)).content}export function estimateTokens(text){return Math.ceil(text.length/4)}export const anthropicDriver={create:createAnthropicDriver},anthropic={configure,chat,streamChat,streamChatEvents,prompt,estimateTokens,createDriver:createAnthropicDriver};export default anthropic;

package/dist/drivers/bedrock/index.d.ts

@@ -1,4 +1,4 @@
-import type { AIMessage, AIResult, ChatCompletionOptions } from '../../types';
+import type { AIMessage, AIResult, AIStreamEvent, ChatCompletionOptions } from '../../types';
 import type { ConverseCommandInput } from '@stacksjs/ts-cloud/aws';
 export declare function configure(config: BedrockDriverConfig): void;
 /**
@@ -10,6 +10,23 @@ export declare function toConverseMessages(messages: AIMessage[], system?: strin
  * One chat completion. `options.model` overrides the configured model.
  */
 export declare function chat(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): Promise<AIResult>;
+/**
+ * A streamed chat completion, as events: text as it is written, each tool
+ * call once its arguments have streamed, then the whole result.
+ *
+ * Converse streams a tool call as a `contentBlockStart` carrying its id and
+ * name, then `toolUse.input` fragments of one JSON string, so a call is
+ * reported at its `contentBlockStop`, whole. The structured-output tool is
+ * internal, as in `chat()`: its JSON streams as text. Usage arrives last, in
+ * `metadata`.
+ */
+export declare function streamChatEvents(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): AsyncGenerator<AIStreamEvent>;
+/**
+ * Stream chat completion: the text as it is written. Tool calls, usage and
+ * the stop reason are in the result the generator returns; iterate
+ * {@link streamChatEvents} to act on a tool call as soon as it arrives.
+ */
+export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions & { system?: string }): AsyncGenerator<string, AIResult, undefined>;
 /**
  * Invocable on demand, with no inference profile and no agreement, so it
  * answers in a fresh account. The Titan Text model this replaces as the
@@ -18,7 +35,9 @@ export declare function chat(messages: AIMessage[], options?: ChatCompletionOpti
 export declare const DEFAULT_BEDROCK_MODEL: 'amazon.nova-lite-v1:0';
 export declare const bedrock: {
   configure: typeof configure;
-  chat: typeof chat
+  chat: typeof chat;
+  streamChat: typeof streamChat;
+  streamChatEvents: typeof streamChatEvents
 };
 export declare interface BedrockDriverConfig {
   model?: string

package/dist/drivers/bedrock/index.js

@@ -1,2 +1,2 @@
-import{converse}from"../../utils/client-bedrock-runtime";import{recordUsage}from"../../utils/usage";export const DEFAULT_BEDROCK_MODEL="amazon.nova-lite-v1:0";let globalConfig={};export function configure(config){globalConfig={...globalConfig,...config}}function resolveConfig(){return{...globalConfig,model:globalConfig.model||process.env.BEDROCK_MODEL_ID||DEFAULT_BEDROCK_MODEL}}const IMAGE_FORMATS=new Set(["png","jpeg","gif","webp"]);function imageBlock(mediaType,data){const format=mediaType.replace(/^image\//,"").replace("jpg","jpeg");if(!IMAGE_FORMATS.has(format))throw Error(`Bedrock Converse takes png, jpeg, gif or webp images, not ${mediaType}`);return{image:{format,source:{bytes:data}}}}function contentBlock(part){if(part.type==="text")return{text:part.text??""};if(part.type==="tool_call"&&part.toolCall)return{toolUse:{toolUseId:part.toolCall.id,name:part.toolCall.name,input:part.toolCall.arguments}};if(part.type==="tool_result"&&part.toolResult)return{toolResult:{toolUseId:part.toolResult.toolCallId,content:[{text:part.toolResult.content}],...part.toolResult.isError?{status:"error"}:{}}};if(part.type==="image"&&part.source)return imageBlock(part.source.media_type,part.source.data);if(part.type==="image_url"&&part.image_url){const inline=/^data:([^;]+);base64,(.*)$/.exec(part.image_url.url);if(!inline)throw Error("Bedrock Converse takes image bytes, not a URL: pass the image as a data: URL or a base64 `image` block");return imageBlock(inline[1],inline[2])}throw Error(`Unsupported message content for Bedrock: ${part.type}`)}export function toConverseMessages(messages,system){const systemBlocks=system?[{text:system}]:[],turns=[];for(const message of messages){if(message.role==="system"){const text=typeof message.content==="string"?message.content:message.content.filter((part)=>part.type==="text").map((part)=>part.text??"").join(`
-`);if(text)systemBlocks.push({text});continue}turns.push({role:message.role,content:typeof message.content==="string"?[{text:message.content}]:message.content.map(contentBlock)})}return{messages:turns,...systemBlocks.length>0?{system:systemBlocks}:{}}}function structuredOutputTool(format){if(format.type==="json_schema")return{name:format.json_schema.name,description:`Returns the result as JSON matching the '${format.json_schema.name}' schema.`,inputSchema:{json:format.json_schema.schema}};return{name:"structured_output",description:"Returns the result as a JSON object.",inputSchema:{json:{type:"object",additionalProperties:!0}}}}function toolConfig(options){const{tools=[],toolChoice,responseFormat}=options,specs=toolChoice==="none"?[]:tools.map((tool)=>({name:tool.name,description:tool.description,inputSchema:{json:tool.parameters??{type:"object",properties:{}}}}));let choice;if(toolChoice==="required")choice={any:{}};else if(toolChoice&&typeof toolChoice==="object")choice={tool:{name:toolChoice.name}};else if(toolChoice==="auto")choice={auto:{}};if(responseFormat&&responseFormat.type!=="text"){const output=structuredOutputTool(responseFormat);specs.push(output);choice=choice??{tool:{name:output.name}}}if(specs.length===0)return;return{tools:specs.map((toolSpec)=>({toolSpec})),...choice?{toolChoice:choice}:{}}}function readOutput(output,outputTool){const blocks=output.output.message?.content??[],uses=blocks.filter((block)=>("toolUse"in block)),structured=outputTool?uses.find((block)=>block.toolUse.name===outputTool):void 0,toolCalls=uses.filter((block)=>block!==structured).map((block)=>({id:block.toolUse.toolUseId,name:block.toolUse.name,arguments:block.toolUse.input??{}}));return{content:structured?JSON.stringify(structured.toolUse.input):blocks.filter((block)=>("text"in block)).map((block)=>block.text).join(""),...toolCalls.length>0?{toolCalls}:{}}}export async function chat(messages,options={}){const config=resolveConfig(),model=options.model||config.model,startedAt=Date.now(),inferenceConfig={},maxTokens=options.maxTokens??config.maxTokens;if(maxTokens!==void 0)inferenceConfig.maxTokens=maxTokens;if(options.temperature!==void 0)inferenceConfig.temperature=options.temperature;if(options.topP!==void 0)inferenceConfig.topP=options.topP;if(options.stop!==void 0)inferenceConfig.stopSequences=Array.isArray(options.stop)?options.stop:[options.stop];const tools=toolConfig(options),response=await converse({modelId:model,...toConverseMessages(messages,options.system),...Object.keys(inferenceConfig).length>0?{inferenceConfig}:{},...tools?{toolConfig:tools}:{}},config.region),outputTool=options.responseFormat&&options.responseFormat.type!=="text"?structuredOutputTool(options.responseFormat).name:void 0,result={...readOutput(response,outputTool),model,usage:{promptTokens:response.usage?.inputTokens??0,completionTokens:response.usage?.outputTokens??0,totalTokens:response.usage?.totalTokens??0},finishReason:response.stopReason};recordUsage({provider:"bedrock",model,promptTokens:result.usage.promptTokens,completionTokens:result.usage.completionTokens,totalTokens:result.usage.totalTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()});return result}export const bedrock={configure,chat};
+import{converse,converseStream}from"../../utils/client-bedrock-runtime";import{streamText}from"../../utils/stream";import{parseToolArguments}from"../../utils/tools";import{recordUsage}from"../../utils/usage";export const DEFAULT_BEDROCK_MODEL="amazon.nova-lite-v1:0";let globalConfig={};export function configure(config){globalConfig={...globalConfig,...config}}function resolveConfig(){return{...globalConfig,model:globalConfig.model||process.env.BEDROCK_MODEL_ID||DEFAULT_BEDROCK_MODEL}}const IMAGE_FORMATS=new Set(["png","jpeg","gif","webp"]);function imageBlock(mediaType,data){const format=mediaType.replace(/^image\//,"").replace("jpg","jpeg");if(!IMAGE_FORMATS.has(format))throw Error(`Bedrock Converse takes png, jpeg, gif or webp images, not ${mediaType}`);return{image:{format,source:{bytes:data}}}}function contentBlock(part){if(part.type==="text")return{text:part.text??""};if(part.type==="tool_call"&&part.toolCall)return{toolUse:{toolUseId:part.toolCall.id,name:part.toolCall.name,input:part.toolCall.arguments}};if(part.type==="tool_result"&&part.toolResult)return{toolResult:{toolUseId:part.toolResult.toolCallId,content:[{text:part.toolResult.content}],...part.toolResult.isError?{status:"error"}:{}}};if(part.type==="image"&&part.source)return imageBlock(part.source.media_type,part.source.data);if(part.type==="image_url"&&part.image_url){const inline=/^data:([^;]+);base64,(.*)$/.exec(part.image_url.url);if(!inline)throw Error("Bedrock Converse takes image bytes, not a URL: pass the image as a data: URL or a base64 `image` block");return imageBlock(inline[1],inline[2])}throw Error(`Unsupported message content for Bedrock: ${part.type}`)}export function toConverseMessages(messages,system){const systemBlocks=system?[{text:system}]:[],turns=[];for(const message of messages){if(message.role==="system"){const text=typeof message.content==="string"?message.content:message.content.filter((part)=>part.type==="text").map((part)=>part.text??"").join(`
+`);if(text)systemBlocks.push({text});continue}turns.push({role:message.role,content:typeof message.content==="string"?[{text:message.content}]:message.content.map(contentBlock)})}return{messages:turns,...systemBlocks.length>0?{system:systemBlocks}:{}}}function structuredOutputTool(format){if(format.type==="json_schema")return{name:format.json_schema.name,description:`Returns the result as JSON matching the '${format.json_schema.name}' schema.`,inputSchema:{json:format.json_schema.schema}};return{name:"structured_output",description:"Returns the result as a JSON object.",inputSchema:{json:{type:"object",additionalProperties:!0}}}}function toolConfig(options){const{tools=[],toolChoice,responseFormat}=options,specs=toolChoice==="none"?[]:tools.map((tool)=>({name:tool.name,description:tool.description,inputSchema:{json:tool.parameters??{type:"object",properties:{}}}}));let choice;if(toolChoice==="required")choice={any:{}};else if(toolChoice&&typeof toolChoice==="object")choice={tool:{name:toolChoice.name}};else if(toolChoice==="auto")choice={auto:{}};if(responseFormat&&responseFormat.type!=="text"){const output=structuredOutputTool(responseFormat);specs.push(output);choice=choice??{tool:{name:output.name}}}if(specs.length===0)return;return{tools:specs.map((toolSpec)=>({toolSpec})),...choice?{toolChoice:choice}:{}}}function readOutput(output,outputTool){const blocks=output.output.message?.content??[],uses=blocks.filter((block)=>("toolUse"in block)),structured=outputTool?uses.find((block)=>block.toolUse.name===outputTool):void 0,toolCalls=uses.filter((block)=>block!==structured).map((block)=>({id:block.toolUse.toolUseId,name:block.toolUse.name,arguments:block.toolUse.input??{}}));return{content:structured?JSON.stringify(structured.toolUse.input):blocks.filter((block)=>("text"in block)).map((block)=>block.text).join(""),...toolCalls.length>0?{toolCalls}:{}}}function converseRequest(messages,options){const config=resolveConfig(),model=options.model||config.model,inferenceConfig={},maxTokens=options.maxTokens??config.maxTokens;if(maxTokens!==void 0)inferenceConfig.maxTokens=maxTokens;if(options.temperature!==void 0)inferenceConfig.temperature=options.temperature;if(options.topP!==void 0)inferenceConfig.topP=options.topP;if(options.stop!==void 0)inferenceConfig.stopSequences=Array.isArray(options.stop)?options.stop:[options.stop];const tools=toolConfig(options),outputTool=options.responseFormat&&options.responseFormat.type!=="text"?structuredOutputTool(options.responseFormat).name:void 0;return{input:{modelId:model,...toConverseMessages(messages,options.system),...Object.keys(inferenceConfig).length>0?{inferenceConfig}:{},...tools?{toolConfig:tools}:{}},...outputTool?{outputTool}:{},region:config.region}}function report(model,usage,startedAt){recordUsage({provider:"bedrock",model,promptTokens:usage.promptTokens,completionTokens:usage.completionTokens,totalTokens:usage.totalTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()})}export async function chat(messages,options={}){const startedAt=Date.now(),{input,outputTool,region}=converseRequest(messages,options),response=await converse(input,region),result={...readOutput(response,outputTool),model:input.modelId,usage:{promptTokens:response.usage?.inputTokens??0,completionTokens:response.usage?.outputTokens??0,totalTokens:response.usage?.totalTokens??0},finishReason:response.stopReason};report(input.modelId,result.usage,startedAt);return result}export async function*streamChatEvents(messages,options={}){const startedAt=Date.now(),{input,outputTool,region}=converseRequest(messages,options),{stream}=await converseStream(input,region),toolBlocks=new Map,toolCalls=[];let content="",usage,finishReason;for await(const event of stream)if("contentBlockStart"in event){const use=event.contentBlockStart.start.toolUse;if(use)toolBlocks.set(event.contentBlockStart.contentBlockIndex,{id:use.toolUseId,name:use.name,json:"",structured:use.name===outputTool})}else if("contentBlockDelta"in event){const{contentBlockIndex,delta}=event.contentBlockDelta;if(delta.text){content+=delta.text;yield{type:"text",text:delta.text}}else if(delta.toolUse){const block=toolBlocks.get(contentBlockIndex);if(!block)continue;block.json+=delta.toolUse.input;if(block.structured&&delta.toolUse.input){content+=delta.toolUse.input;yield{type:"text",text:delta.toolUse.input}}}}else if("contentBlockStop"in event){const block=toolBlocks.get(event.contentBlockStop.contentBlockIndex);if(block&&!block.structured){const call={id:block.id,name:block.name,arguments:parseToolArguments(block.json,block.name)};toolCalls.push(call);yield{type:"tool_call",call}}}else if("messageStop"in event)finishReason=event.messageStop.stopReason;else if("metadata"in event)usage={promptTokens:event.metadata.usage.inputTokens,completionTokens:event.metadata.usage.outputTokens,totalTokens:event.metadata.usage.totalTokens};if(usage)report(input.modelId,usage,startedAt);yield{type:"done",result:{content,...toolCalls.length>0?{toolCalls}:{},model:input.modelId,...usage?{usage}:{},...finishReason?{finishReason}:{}}}}export function streamChat(messages,options={}){return streamText(streamChatEvents(messages,options))}export const bedrock={configure,chat,streamChat,streamChatEvents};

package/dist/drivers/ollama/index.d.ts

@@ -1,4 +1,4 @@
-import type { AIDriver, AIDriverConfig, AIMessage, AIResult, ChatCompletionOptions } from '../../types';
+import type { AIDriver, AIDriverConfig, AIMessage, AIResult, AIStreamEvent, ChatCompletionOptions } from '../../types';
 /**
  * Configure Ollama globally
  */
@@ -9,9 +9,20 @@ export declare function createOllamaDriver(config?: OllamaDriverConfig): AIDrive
  */
 export declare function chat(messages: AIMessage[], options?: ChatCompletionOptions): Promise<AIResult>;
 /**
- * Stream chat completion
+ * A streamed chat completion, as events: text as it is written, each tool
+ * call as it arrives, then the whole result.
+ *
+ * Ollama streams newline-delimited JSON. A tool call arrives whole, on the
+ * message of whichever chunk carries it, and the counts and stop reason on
+ * the last (`done: true`) one. An `error` line throws.
  */
-export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions): AsyncGenerator<string>;
+export declare function streamChatEvents(messages: AIMessage[], options?: ChatCompletionOptions): AsyncGenerator<AIStreamEvent>;
+/**
+ * Stream chat completion: the text as it is written. Tool calls, usage and
+ * the stop reason are in the result the generator returns; iterate
+ * {@link streamChatEvents} to act on a tool call as soon as it arrives.
+ */
+export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions): AsyncGenerator<string, AIResult, undefined>;
 /**
  * Generate text completion (non-chat)
  */
@@ -76,6 +87,7 @@ export declare const ollama: {
   configure: typeof configure;
   chat: typeof chat;
   streamChat: typeof streamChat;
+  streamChatEvents: typeof streamChatEvents;
   generate: typeof generate;
   embed: typeof embed;
   listModels: typeof listModels;
@@ -90,4 +102,5 @@ export declare interface OllamaDriverConfig extends AIDriverConfig {
   model?: string
   embeddingModel?: string
 }
+declare type OllamaToolCalls = Array<{ id?: string, function: { name: string, arguments: unknown } }>;
 export default ollama;

package/dist/drivers/ollama/index.js

@@ -1,4 +1,2 @@
-import{parseToolArguments,toOllamaMessages}from"../../utils/tools";const DEFAULT_HOST="http://localhost:11434",DEFAULT_MODEL="llama3.2",DEFAULT_EMBEDDING_MODEL="nomic-embed-text";let globalConfig={};export function configure(config){globalConfig={...globalConfig,...config}}function getConfig(config){return{host:config?.host||globalConfig.host||process.env.OLLAMA_HOST||DEFAULT_HOST,model:config?.model||globalConfig.model||process.env.OLLAMA_MODEL||DEFAULT_MODEL,embeddingModel:config?.embeddingModel||globalConfig.embeddingModel||DEFAULT_EMBEDDING_MODEL}}export function createOllamaDriver(config={}){const{host=process.env.OLLAMA_HOST||DEFAULT_HOST,model=process.env.OLLAMA_MODEL||DEFAULT_MODEL,embeddingModel=DEFAULT_EMBEDDING_MODEL}=config;return{name:"Ollama",async process(command,systemPrompt,history){const response=await fetch(`${host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}],stream:!1})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}return(await response.json()).message.content},async*stream(command,systemPrompt,history){const response=await fetch(`${host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}],stream:!0})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const reader=response.body?.getReader();if(!reader)throw Error("No response body");const decoder=new TextDecoder;let buffer="";while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines){if(!line.trim())continue;try{const data=JSON.parse(line);if(data.message?.content)yield data.message.content}catch{}}}},async embed(input){const inputs=Array.isArray(input)?input:[input],embeddings=[];for(const text of inputs){const response=await fetch(`${host}/api/embeddings`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model:embeddingModel,prompt:text})});if(!response.ok){const error=await response.text();throw Error(`Ollama Embeddings API error: ${error}`)}const data=await response.json();embeddings.push(data.embedding)}return Array.isArray(input)?embeddings:embeddings[0]}}}export async function chat(messages,options={}){const config=getConfig(),{model=config.model,temperature,topP,stop,responseFormat,tools,toolChoice}=options;if(toolChoice==="required"||toolChoice&&typeof toolChoice==="object")throw Error("Ollama cannot be made to call a tool: offer the tools with toolChoice 'auto', or use another driver");const offered=toolChoice==="none"?[]:tools??[],format=responseFormat?.type==="json_schema"?responseFormat.json_schema.schema:responseFormat?.type==="json_object"?"json":void 0,response=await fetch(`${config.host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model,messages:toOllamaMessages(messages),stream:!1,format,...offered.length>0?{tools:offered.map((tool)=>({type:"function",function:{name:tool.name,description:tool.description,parameters:tool.parameters??{type:"object",properties:{}}}}))}:{},options:{temperature,top_p:topP,stop}})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const data=await response.json(),toolCalls=(data.message?.tool_calls??[]).map((call,index)=>({id:call.id??`call_${index}`,name:call.function.name,arguments:parseToolArguments(call.function.arguments,call.function.name)}));return{content:data.message?.content??"",...toolCalls.length>0?{toolCalls}:{},model:data.model,usage:{promptTokens:data.prompt_eval_count||0,completionTokens:data.eval_count||0,totalTokens:(data.prompt_eval_count||0)+(data.eval_count||0)},finishReason:data.done_reason||"stop"}}export async function*streamChat(messages,options={}){const config=getConfig(),{model=config.model,temperature,topP,stop}=options,response=await fetch(`${config.host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model,messages:toOllamaMessages(messages),stream:!0,options:{temperature,top_p:topP,stop}})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const reader=response.body?.getReader();if(!reader)throw Error("No response body");const decoder=new TextDecoder;let buffer="";while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines){if(!line.trim())continue;try{const data=JSON.parse(line);if(data.message?.content)yield data.message.content}catch{}}}}export async function generate(prompt,options={}){const config=getConfig(),response=await fetch(`${config.host}/api/generate`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model:options.model||config.model,prompt,system:options.system,template:options.template,context:options.context,raw:options.raw,format:options.format,images:options.images,stream:!1})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const data=await response.json();return{content:data.response,model:data.model,usage:{promptTokens:data.prompt_eval_count||0,completionTokens:data.eval_count||0,totalTokens:(data.prompt_eval_count||0)+(data.eval_count||0)}}}export async function embed(input,model){const config=getConfig(),embeddingModel=model||config.embeddingModel,inputs=Array.isArray(input)?input:[input],embeddings=[];for(const text of inputs){const response=await fetch(`${config.host}/api/embeddings`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model:embeddingModel,prompt:text})});if(!response.ok){const error=await response.text();throw Error(`Ollama Embeddings API error: ${error}`)}const data=await response.json();embeddings.push(data.embedding)}return Array.isArray(input)?embeddings:embeddings[0]}export async function listModels(){const config=getConfig(),response=await fetch(`${config.host}/api/tags`,{method:"GET"});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}return(await response.json()).models}export async function pullModel(name,onProgress){const config=getConfig(),response=await fetch(`${config.host}/api/pull`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({name,stream:!0})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const reader=response.body?.getReader();if(!reader)return;const decoder=new TextDecoder;let buffer="";while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines){if(!line.trim())continue;try{const data=JSON.parse(line);if(onProgress)onProgress(data.status??"",data.completed,data.total)}catch{}}}}export async function deleteModel(name){const config=getConfig(),response=await fetch(`${config.host}/api/delete`,{method:"DELETE",headers:{"Content-Type":"application/json"},body:JSON.stringify({name})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}}export async function showModel(_name){const config=getConfig(),response=await fetch(`${config.host}/api/show`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({name:_name})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}return response.json()}export async function isRunning(){const config=getConfig();try{return(await fetch(`${config.host}/api/tags`,{method:"GET"})).ok}catch{return!1}}export const ollamaDriver={create:createOllamaDriver},ollama={configure,chat,streamChat,generate,embed,listModels,pullModel,deleteModel,showModel,isRunning,createDriver:createOllamaDriver};export default ollama;
+import{ndjson,streamText}from"../../utils/stream";import{parseToolArguments,toOllamaMessages}from"../../utils/tools";const DEFAULT_HOST="http://localhost:11434",DEFAULT_MODEL="llama3.2",DEFAULT_EMBEDDING_MODEL="nomic-embed-text";let globalConfig={};export function configure(config){globalConfig={...globalConfig,...config}}function getConfig(config){return{host:config?.host||globalConfig.host||process.env.OLLAMA_HOST||DEFAULT_HOST,model:config?.model||globalConfig.model||process.env.OLLAMA_MODEL||DEFAULT_MODEL,embeddingModel:config?.embeddingModel||globalConfig.embeddingModel||DEFAULT_EMBEDDING_MODEL}}export function createOllamaDriver(config={}){const{host=process.env.OLLAMA_HOST||DEFAULT_HOST,model=process.env.OLLAMA_MODEL||DEFAULT_MODEL,embeddingModel=DEFAULT_EMBEDDING_MODEL}=config;return{name:"Ollama",async process(command,systemPrompt,history){const response=await fetch(`${host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}],stream:!1})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}return(await response.json()).message.content},async*stream(command,systemPrompt,history){const response=await fetch(`${host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}],stream:!0})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}if(!response.body)throw Error("No response body");for await(const data of ndjson(response.body,"Ollama API")){if(data.error)throw Error(`Ollama API error: ${data.error}`);if(data.message?.content)yield data.message.content}},async embed(input){const inputs=Array.isArray(input)?input:[input],embeddings=[];for(const text of inputs){const response=await fetch(`${host}/api/embeddings`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model:embeddingModel,prompt:text})});if(!response.ok){const error=await response.text();throw Error(`Ollama Embeddings API error: ${error}`)}const data=await response.json();embeddings.push(data.embedding)}return Array.isArray(input)?embeddings:embeddings[0]}}}function chatRequest(messages,options,stream){const config=getConfig(),{model=config.model,temperature,topP,stop,responseFormat,tools,toolChoice}=options;if(toolChoice==="required"||toolChoice&&typeof toolChoice==="object")throw Error("Ollama cannot be made to call a tool: offer the tools with toolChoice 'auto', or use another driver");const offered=toolChoice==="none"?[]:tools??[],format=responseFormat?.type==="json_schema"?responseFormat.json_schema.schema:responseFormat?.type==="json_object"?"json":void 0;return{model,messages:toOllamaMessages(messages),stream,format,...offered.length>0?{tools:offered.map((tool)=>({type:"function",function:{name:tool.name,description:tool.description,parameters:tool.parameters??{type:"object",properties:{}}}}))}:{},options:{temperature,top_p:topP,stop}}}function toToolCalls(calls,offset=0){return calls.map((call,index)=>({id:call.id??`call_${offset+index}`,name:call.function.name,arguments:parseToolArguments(call.function.arguments,call.function.name)}))}export async function chat(messages,options={}){const config=getConfig(),response=await fetch(`${config.host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify(chatRequest(messages,options,!1))});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const data=await response.json(),toolCalls=toToolCalls(data.message?.tool_calls??[]);return{content:data.message?.content??"",...toolCalls.length>0?{toolCalls}:{},model:data.model,usage:{promptTokens:data.prompt_eval_count||0,completionTokens:data.eval_count||0,totalTokens:(data.prompt_eval_count||0)+(data.eval_count||0)},finishReason:data.done_reason||"stop"}}export async function*streamChatEvents(messages,options={}){const config=getConfig(),body=chatRequest(messages,options,!0),response=await fetch(`${config.host}/api/chat`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify(body)});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}if(!response.body)throw Error("No response body");const toolCalls=[];let content="",model=String(body.model),usage,finishReason;for await(const chunk of ndjson(response.body,"Ollama API")){if(chunk.error)throw Error(`Ollama API error: ${chunk.error}`);if(chunk.model)model=chunk.model;if(chunk.message?.content){content+=chunk.message.content;yield{type:"text",text:chunk.message.content}}for(const call of toToolCalls(chunk.message?.tool_calls??[],toolCalls.length)){toolCalls.push(call);yield{type:"tool_call",call}}if(chunk.done){finishReason=chunk.done_reason||"stop";usage={promptTokens:chunk.prompt_eval_count||0,completionTokens:chunk.eval_count||0,totalTokens:(chunk.prompt_eval_count||0)+(chunk.eval_count||0)}}}yield{type:"done",result:{content,...toolCalls.length>0?{toolCalls}:{},model,...usage?{usage}:{},...finishReason?{finishReason}:{}}}}export function streamChat(messages,options={}){return streamText(streamChatEvents(messages,options))}export async function generate(prompt,options={}){const config=getConfig(),response=await fetch(`${config.host}/api/generate`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model:options.model||config.model,prompt,system:options.system,template:options.template,context:options.context,raw:options.raw,format:options.format,images:options.images,stream:!1})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const data=await response.json();return{content:data.response,model:data.model,usage:{promptTokens:data.prompt_eval_count||0,completionTokens:data.eval_count||0,totalTokens:(data.prompt_eval_count||0)+(data.eval_count||0)}}}export async function embed(input,model){const config=getConfig(),embeddingModel=model||config.embeddingModel,inputs=Array.isArray(input)?input:[input],embeddings=[];for(const text of inputs){const response=await fetch(`${config.host}/api/embeddings`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({model:embeddingModel,prompt:text})});if(!response.ok){const error=await response.text();throw Error(`Ollama Embeddings API error: ${error}`)}const data=await response.json();embeddings.push(data.embedding)}return Array.isArray(input)?embeddings:embeddings[0]}export async function listModels(){const config=getConfig(),response=await fetch(`${config.host}/api/tags`,{method:"GET"});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}return(await response.json()).models}export async function pullModel(name,onProgress){const config=getConfig(),response=await fetch(`${config.host}/api/pull`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({name,stream:!0})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}const reader=response.body?.getReader();if(!reader)return;const decoder=new TextDecoder;let buffer="";while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
+`);buffer=lines.pop()||"";for(const line of lines){if(!line.trim())continue;try{const data=JSON.parse(line);if(onProgress)onProgress(data.status??"",data.completed,data.total)}catch{}}}}export async function deleteModel(name){const config=getConfig(),response=await fetch(`${config.host}/api/delete`,{method:"DELETE",headers:{"Content-Type":"application/json"},body:JSON.stringify({name})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}}export async function showModel(_name){const config=getConfig(),response=await fetch(`${config.host}/api/show`,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({name:_name})});if(!response.ok){const error=await response.text();throw Error(`Ollama API error: ${error}`)}return response.json()}export async function isRunning(){const config=getConfig();try{return(await fetch(`${config.host}/api/tags`,{method:"GET"})).ok}catch{return!1}}export const ollamaDriver={create:createOllamaDriver},ollama={configure,chat,streamChat,streamChatEvents,generate,embed,listModels,pullModel,deleteModel,showModel,isRunning,createDriver:createOllamaDriver};export default ollama;

package/dist/drivers/openai/index.d.ts

@@ -1,4 +1,4 @@
-import type { AIDriver, AIDriverConfig, AIMessage, AIResult, ChatCompletionOptions } from '../../types';
+import type { AIDriver, AIDriverConfig, AIMessage, AIResult, AIStreamEvent, ChatCompletionOptions } from '../../types';
 /**
  * Configure OpenAI globally
  */
@@ -9,9 +9,22 @@ export declare function createOpenAIDriver(config: OpenAIDriverConfig): AIDriver
  */
 export declare function chat(messages: AIMessage[], options?: ChatCompletionOptions): Promise<AIResult>;
 /**
- * Stream chat completion
+ * A streamed chat completion, as events: text as it is written, each tool
+ * call once its arguments have streamed, then the whole result.
+ *
+ * A tool call streams as `delta.tool_calls` pieces keyed by `index`: the id
+ * and name first, then the arguments a fragment at a time. The calls are
+ * complete when the choice finishes, and are reported then, in index order.
+ * Usage only comes with `stream_options.include_usage`, on a last chunk with
+ * no choices, so it is asked for. An `error` chunk throws.
  */
-export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions): AsyncGenerator<string>;
+export declare function streamChatEvents(messages: AIMessage[], options?: ChatCompletionOptions): AsyncGenerator<AIStreamEvent>;
+/**
+ * Stream chat completion: the text as it is written. Tool calls, usage and
+ * the stop reason are in the result the generator returns; iterate
+ * {@link streamChatEvents} to act on a tool call as soon as it arrives.
+ */
+export declare function streamChat(messages: AIMessage[], options?: ChatCompletionOptions): AsyncGenerator<string, AIResult, undefined>;
 /**
  * Create embeddings
  */
@@ -50,6 +63,7 @@ export declare const openai: {
   configure: typeof configure;
   chat: typeof chat;
   streamChat: typeof streamChat;
+  streamChatEvents: typeof streamChatEvents;
   embed: typeof embed;
   generateImage: typeof generateImage;
   transcribe: typeof transcribe;

package/dist/drivers/openai/index.js

@@ -1,3 +1 @@
-import{fetchWithRetry}from"../../utils/retry";import{parseToolArguments,toChatCompletionMessages}from"../../utils/tools";import{recordUsage}from"../../utils/usage";const DEFAULT_MODEL="gpt-4o",DEFAULT_MAX_TOKENS=4096,DEFAULT_EMBEDDING_MODEL="text-embedding-3-small",DEFAULT_BASE_URL="https://api.openai.com/v1";let globalConfig=null;export function configure(config){globalConfig=config}function getConfig(config){const merged={...globalConfig,...config};if(!merged.apiKey)merged.apiKey=process.env.OPENAI_API_KEY||"";return merged}export function createOpenAIDriver(config){const{apiKey,model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,baseUrl=DEFAULT_BASE_URL,embeddingModel=DEFAULT_EMBEDDING_MODEL}=config;return{name:"OpenAI",async process(command,systemPrompt,history){if(!apiKey)throw Error("OpenAI API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${baseUrl}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${apiKey}`},body:JSON.stringify({model,max_tokens:maxTokens,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}const data=await response.json();if(!data.choices||data.choices.length===0)throw Error("OpenAI API returned empty choices");return data.choices[0].message.content},async*stream(command,systemPrompt,history){if(!apiKey)throw Error("OpenAI API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${baseUrl}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${apiKey}`},body:JSON.stringify({model,max_tokens:maxTokens,stream:!0,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}const reader=response.body?.getReader();if(!reader)throw Error("No response body");const decoder=new TextDecoder;let buffer="";const handlePayload=function*(data){if(data==="[DONE]")return;let parsed;try{parsed=JSON.parse(data)}catch{return}if(parsed?.error){const msg=parsed.error.message||JSON.stringify(parsed.error);throw Error(`[openai/stream] mid-stream error: ${msg}`)}const content=parsed.choices?.[0]?.delta?.content;if(content)yield content};while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines)if(line.startsWith("data: "))yield*handlePayload(line.slice(6))}if(buffer.startsWith("data: "))yield*handlePayload(buffer.slice(6))},async embed(input){if(!apiKey)throw Error("OpenAI API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${baseUrl}/embeddings`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${apiKey}`},body:JSON.stringify({model:embeddingModel,input})});if(!response.ok){const error=await response.text();throw Error(`OpenAI Embeddings API error: ${error}`)}const data=await response.json();if(Array.isArray(input))return data.data.map((d)=>d.embedding);return data.data[0].embedding}}}export async function chat(messages,options={}){const config=getConfig(),{model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,temperature,topP,stop,tools,toolChoice,responseFormat}=options,normalizedMessages=toChatCompletionMessages(messages,{argumentsAsString:!0}),startedAt=Date.now(),body={model,max_tokens:maxTokens,temperature,top_p:topP,stop,messages:normalizedMessages};if(tools?.length)body.tools=tools.map((tool)=>({type:"function",function:{name:tool.name,description:tool.description,parameters:tool.parameters??{type:"object",properties:{}}}}));if(toolChoice!==void 0)body.tool_choice=typeof toolChoice==="object"?{type:"function",function:{name:toolChoice.name}}:toolChoice;if(responseFormat&&responseFormat.type!=="text")body.response_format=responseFormat;const response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify(body)});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}const data=await response.json();if(!data.choices||data.choices.length===0)throw Error("OpenAI API returned empty choices");const message=data.choices[0].message,toolCalls=(message.tool_calls??[]).map((call)=>({id:call.id,name:call.function.name,arguments:parseToolArguments(call.function.arguments,call.function.name)})),result={content:message.content??"",...toolCalls.length>0?{toolCalls}:{},model:data.model,usage:{promptTokens:data.usage?.prompt_tokens||0,completionTokens:data.usage?.completion_tokens||0,totalTokens:data.usage?.total_tokens||0},finishReason:data.choices[0].finish_reason};recordUsage({provider:"openai",model:data.model,promptTokens:result.usage.promptTokens,completionTokens:result.usage.completionTokens,totalTokens:result.usage.totalTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()});return result}export async function*streamChat(messages,options={}){const config=getConfig(),{model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,temperature,topP,stop}=options,response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,max_tokens:maxTokens,temperature,top_p:topP,stop,stream:!0,messages:toChatCompletionMessages(messages,{argumentsAsString:!0})})});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}const reader=response.body?.getReader();if(!reader)throw Error("No response body");const decoder=new TextDecoder;let buffer="";while(!0){const{done,value}=await reader.read();if(done)break;buffer+=decoder.decode(value,{stream:!0});const lines=buffer.split(`
-`);buffer=lines.pop()||"";for(const line of lines)if(line.startsWith("data: ")){const data=line.slice(6);if(data==="[DONE]")continue;try{const content=JSON.parse(data).choices[0]?.delta?.content;if(content)yield content}catch{}}}}export async function embed(input,model=DEFAULT_EMBEDDING_MODEL){const config=getConfig(),response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/embeddings`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,input})});if(!response.ok){const error=await response.text();throw Error(`OpenAI Embeddings API error: ${error}`)}const data=await response.json();if(Array.isArray(input))return data.data.map((d)=>d.embedding);return data.data[0].embedding}export async function generateImage(prompt,options={}){const config=getConfig(),{model="dall-e-3",size="1024x1024",quality="standard",n=1,responseFormat="url"}=options,response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/images/generations`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,prompt,size,quality,n,response_format:responseFormat})});if(!response.ok){const error=await response.text();throw Error(`OpenAI Image API error: ${error}`)}return(await response.json()).data}export async function transcribe(audioFile,options={}){const config=getConfig(),formData=new FormData;formData.append("file",audioFile);formData.append("model",options.model||"whisper-1");if(options.language)formData.append("language",options.language);if(options.prompt)formData.append("prompt",options.prompt);if(options.responseFormat)formData.append("response_format",options.responseFormat);if(options.temperature!==void 0)formData.append("temperature",String(options.temperature));const response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/audio/transcriptions`,{method:"POST",headers:{Authorization:`Bearer ${config.apiKey}`},body:formData});if(!response.ok){const error=await response.text();throw Error(`OpenAI Whisper API error: ${error}`)}return response.json()}export async function textToSpeech(input,options={}){const config=getConfig(),{model="tts-1",voice="alloy",responseFormat="mp3",speed=1}=options,response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/audio/speech`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,input,voice,response_format:responseFormat,speed})});if(!response.ok){const error=await response.text();throw Error(`OpenAI TTS API error: ${error}`)}return response.arrayBuffer()}export const openaiDriver={create:createOpenAIDriver},openai={configure,chat,streamChat,embed,generateImage,transcribe,textToSpeech,createDriver:createOpenAIDriver};export default openai;
+import{fetchWithRetry}from"../../utils/retry";import{sseJson,streamText}from"../../utils/stream";import{parseToolArguments,toChatCompletionMessages}from"../../utils/tools";import{recordUsage}from"../../utils/usage";const DEFAULT_MODEL="gpt-4o",DEFAULT_MAX_TOKENS=4096,DEFAULT_EMBEDDING_MODEL="text-embedding-3-small",DEFAULT_BASE_URL="https://api.openai.com/v1";let globalConfig=null;export function configure(config){globalConfig=config}function getConfig(config){const merged={...globalConfig,...config};if(!merged.apiKey)merged.apiKey=process.env.OPENAI_API_KEY||"";return merged}export function createOpenAIDriver(config){const{apiKey,model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,baseUrl=DEFAULT_BASE_URL,embeddingModel=DEFAULT_EMBEDDING_MODEL}=config;return{name:"OpenAI",async process(command,systemPrompt,history){if(!apiKey)throw Error("OpenAI API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${baseUrl}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${apiKey}`},body:JSON.stringify({model,max_tokens:maxTokens,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}const data=await response.json();if(!data.choices||data.choices.length===0)throw Error("OpenAI API returned empty choices");return data.choices[0].message.content},async*stream(command,systemPrompt,history){if(!apiKey)throw Error("OpenAI API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${baseUrl}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${apiKey}`},body:JSON.stringify({model,max_tokens:maxTokens,stream:!0,messages:[{role:"system",content:systemPrompt},...history,{role:"user",content:command}]})});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}if(!response.body)throw Error("No response body");for await(const parsed of sseJson(response.body,"OpenAI API")){if(parsed?.error){const msg=parsed.error.message||JSON.stringify(parsed.error);throw Error(`[openai/stream] mid-stream error: ${msg}`)}const content=parsed.choices?.[0]?.delta?.content;if(content)yield content}},async embed(input){if(!apiKey)throw Error("OpenAI API key not set. Configure your API key in settings.");const response=await fetchWithRetry(`${baseUrl}/embeddings`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${apiKey}`},body:JSON.stringify({model:embeddingModel,input})});if(!response.ok){const error=await response.text();throw Error(`OpenAI Embeddings API error: ${error}`)}const data=await response.json();if(Array.isArray(input))return data.data.map((d)=>d.embedding);return data.data[0].embedding}}}function completionRequest(messages,options){const{model=DEFAULT_MODEL,maxTokens=DEFAULT_MAX_TOKENS,temperature,topP,stop,tools,toolChoice,responseFormat}=options,body={model,max_tokens:maxTokens,temperature,top_p:topP,stop,messages:toChatCompletionMessages(messages,{argumentsAsString:!0})};if(tools?.length)body.tools=tools.map((tool)=>({type:"function",function:{name:tool.name,description:tool.description,parameters:tool.parameters??{type:"object",properties:{}}}}));if(toolChoice!==void 0)body.tool_choice=typeof toolChoice==="object"?{type:"function",function:{name:toolChoice.name}}:toolChoice;if(responseFormat&&responseFormat.type!=="text")body.response_format=responseFormat;return body}function postCompletion(body){const config=getConfig();return fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/chat/completions`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify(body)})}export async function chat(messages,options={}){const startedAt=Date.now(),response=await postCompletion(completionRequest(messages,options));if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}const data=await response.json();if(!data.choices||data.choices.length===0)throw Error("OpenAI API returned empty choices");const message=data.choices[0].message,toolCalls=(message.tool_calls??[]).map((call)=>({id:call.id,name:call.function.name,arguments:parseToolArguments(call.function.arguments,call.function.name)})),result={content:message.content??"",...toolCalls.length>0?{toolCalls}:{},model:data.model,usage:{promptTokens:data.usage?.prompt_tokens||0,completionTokens:data.usage?.completion_tokens||0,totalTokens:data.usage?.total_tokens||0},finishReason:data.choices[0].finish_reason};recordUsage({provider:"openai",model:data.model,promptTokens:result.usage.promptTokens,completionTokens:result.usage.completionTokens,totalTokens:result.usage.totalTokens,durationMs:Date.now()-startedAt,timestamp:Date.now()});return result}export async function*streamChatEvents(messages,options={}){const startedAt=Date.now(),body=completionRequest(messages,options),response=await postCompletion({...body,stream:!0,stream_options:{include_usage:!0}});if(!response.ok){const error=await response.text();throw Error(`OpenAI API error: ${error}`)}if(!response.body)throw Error("No response body");const pending=new Map,toolCalls=[];let content="",model=String(body.model),usage,finishReason;function*flushCalls(){for(const index of[...pending.keys()].sort((a,b)=>a-b)){const piece=pending.get(index),call={id:piece.id,name:piece.name,arguments:parseToolArguments(piece.args,piece.name)};toolCalls.push(call);yield{type:"tool_call",call}}pending.clear()}for await(const chunk of sseJson(response.body,"OpenAI API")){if(chunk.error)throw Error(`OpenAI API error: ${chunk.error.type?`${chunk.error.type}: `:""}${chunk.error.message??"stream error"}`);if(chunk.model)model=chunk.model;if(chunk.usage)usage={promptTokens:chunk.usage.prompt_tokens??0,completionTokens:chunk.usage.completion_tokens??0,totalTokens:chunk.usage.total_tokens??0};const choice=chunk.choices?.[0];if(!choice)continue;if(choice.delta?.content){content+=choice.delta.content;yield{type:"text",text:choice.delta.content}}for(const piece of choice.delta?.tool_calls??[]){const entry=pending.get(piece.index)??{id:"",name:"",args:""};if(piece.id)entry.id=piece.id;if(piece.function?.name)entry.name+=piece.function.name;if(piece.function?.arguments)entry.args+=piece.function.arguments;pending.set(piece.index,entry)}if(choice.finish_reason){finishReason=choice.finish_reason;yield*flushCalls()}}yield*flushCalls();const result={content,...toolCalls.length>0?{toolCalls}:{},model,...usage?{usage}:{},...finishReason?{finishReason}:{}};if(usage)recordUsage({provider:"openai",model,...usage,durationMs:Date.now()-startedAt,timestamp:Date.now()});yield{type:"done",result}}export function streamChat(messages,options={}){return streamText(streamChatEvents(messages,options))}export async function embed(input,model=DEFAULT_EMBEDDING_MODEL){const config=getConfig(),response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/embeddings`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,input})});if(!response.ok){const error=await response.text();throw Error(`OpenAI Embeddings API error: ${error}`)}const data=await response.json();if(Array.isArray(input))return data.data.map((d)=>d.embedding);return data.data[0].embedding}export async function generateImage(prompt,options={}){const config=getConfig(),{model="dall-e-3",size="1024x1024",quality="standard",n=1,responseFormat="url"}=options,response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/images/generations`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,prompt,size,quality,n,response_format:responseFormat})});if(!response.ok){const error=await response.text();throw Error(`OpenAI Image API error: ${error}`)}return(await response.json()).data}export async function transcribe(audioFile,options={}){const config=getConfig(),formData=new FormData;formData.append("file",audioFile);formData.append("model",options.model||"whisper-1");if(options.language)formData.append("language",options.language);if(options.prompt)formData.append("prompt",options.prompt);if(options.responseFormat)formData.append("response_format",options.responseFormat);if(options.temperature!==void 0)formData.append("temperature",String(options.temperature));const response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/audio/transcriptions`,{method:"POST",headers:{Authorization:`Bearer ${config.apiKey}`},body:formData});if(!response.ok){const error=await response.text();throw Error(`OpenAI Whisper API error: ${error}`)}return response.json()}export async function textToSpeech(input,options={}){const config=getConfig(),{model="tts-1",voice="alloy",responseFormat="mp3",speed=1}=options,response=await fetchWithRetry(`${config.baseUrl||DEFAULT_BASE_URL}/audio/speech`,{method:"POST",headers:{"Content-Type":"application/json",Authorization:`Bearer ${config.apiKey}`},body:JSON.stringify({model,input,voice,response_format:responseFormat,speed})});if(!response.ok){const error=await response.text();throw Error(`OpenAI TTS API error: ${error}`)}return response.arrayBuffer()}export const openaiDriver={create:createOpenAIDriver},openai={configure,chat,streamChat,streamChatEvents,embed,generateImage,transcribe,textToSpeech,createDriver:createOpenAIDriver};export default openai;

package/dist/types.d.ts

@@ -234,6 +234,18 @@ export type AIResponseFormat = | { type: 'text' }
         strict?: boolean
       }
     }
+/**
+ * One event of a streamed chat completion, the same from every driver.
+ *
+ * `text` arrives in pieces as the model writes. A `tool_call` arrives whole,
+ * once its arguments have finished streaming, so it can be run as soon as
+ * it is seen. `done` comes last, exactly once, with everything the stream
+ * carried in the shape `chat()` returns - text, tool calls, usage and why
+ * the model stopped.
+ */
+export type AIStreamEvent = | { type: 'text', text: string }
+  | { type: 'tool_call', call: AIToolCall }
+  | { type: 'done', result: AIResult }
 export type AIProvider = 'anthropic' | 'openai' | 'ollama' | 'bedrock';
 /**
  * Where a provider's credentials come from. `aws-credential-chain`: Bedrock

package/dist/utils/client-bedrock-runtime.d.ts

@@ -1,4 +1,4 @@
-import type { ConverseCommandInput, ConverseCommandOutput, InvokeModelCommandInput, InvokeModelCommandOutput, InvokeModelWithResponseStreamCommandInput, InvokeModelWithResponseStreamCommandOutput } from '@stacksjs/ts-cloud/aws';
+import type { ConverseCommandInput, ConverseCommandOutput, ConverseStreamCommandOutput, InvokeModelCommandInput, InvokeModelCommandOutput, InvokeModelWithResponseStreamCommandInput, InvokeModelWithResponseStreamCommandOutput } from '@stacksjs/ts-cloud/aws';
 export type { ConverseCommandInput, ConverseCommandOutput, InvokeModelCommandInput, InvokeModelWithResponseStreamCommandInput };
 /**
  * The region Bedrock is called in: the one passed, then `AWS_REGION`, then
@@ -13,6 +13,11 @@ export declare function bedrockRegion(region?: string): string;
  * @see https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html
  */
 export declare function converse(params: ConverseCommandInput, region?: string): Promise<ConverseCommandOutput>;
+/*
+ * Converse, streamed: the same request, answered as events
+ * @see https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ConverseStream.html
+ */
+export declare function converseStream(params: ConverseCommandInput, region?: string): Promise<ConverseStreamCommandOutput>;
 /*
  * Invoke Model
  * @param {InvokeModelCommandInput} params

package/dist/utils/client-bedrock-runtime.js

@@ -1 +1 @@
-import process from"node:process";export function bedrockRegion(region){return region||process.env.AWS_REGION||process.env.AWS_DEFAULT_REGION||"us-east-1"}const clients=new Map;async function getClient(region){const resolved=bedrockRegion(region),cached=clients.get(resolved);if(cached)return cached;const mod=await import("@stacksjs/ts-cloud/aws");if(!mod?.BedrockRuntimeClient)throw Error("@stacksjs/ts-cloud/aws does not export BedrockRuntimeClient - rebuild ts-cloud or remove the AI dependency.");const client=new mod.BedrockRuntimeClient(resolved);clients.set(resolved,client);return client}export async function converse(params,region){return(await getClient(region)).converse(params)}export async function invokeModel(params){return(await getClient()).invokeModel(params)}export async function invokeModelWithResponseStream(params){return(await getClient()).invokeModelWithResponseStream(params)}
+import process from"node:process";export function bedrockRegion(region){return region||process.env.AWS_REGION||process.env.AWS_DEFAULT_REGION||"us-east-1"}const clients=new Map;async function getClient(region){const resolved=bedrockRegion(region),cached=clients.get(resolved);if(cached)return cached;const mod=await import("@stacksjs/ts-cloud/aws");if(!mod?.BedrockRuntimeClient)throw Error("@stacksjs/ts-cloud/aws does not export BedrockRuntimeClient - rebuild ts-cloud or remove the AI dependency.");const client=new mod.BedrockRuntimeClient(resolved);clients.set(resolved,client);return client}export async function converse(params,region){return(await getClient(region)).converse(params)}export async function converseStream(params,region){const c=await getClient(region);if(typeof c.converseStream!=="function")throw Error("This @stacksjs/ts-cloud cannot stream the Converse API: upgrade it to 0.16.44 or newer.");return c.converseStream(params)}export async function invokeModel(params){return(await getClient()).invokeModel(params)}export async function invokeModelWithResponseStream(params){return(await getClient()).invokeModelWithResponseStream(params)}

package/dist/utils/stream.d.ts

@@ -0,0 +1,29 @@
+import type { AIResult, AIStreamEvent } from '../types';
+/**
+ * Plumbing shared by the drivers' streamed chat completions.
+ *
+ * Each driver parses its provider's wire format into {@link AIStreamEvent}s
+ * once, in `streamChatEvents()`. `streamChat()` is the text-only view of that
+ * same stream, so a caller that only wants the words keeps writing
+ * `for await (const text of streamChat(...))` and nothing about it changed -
+ * while a tool call the model makes mid-stream, which used to vanish, is in
+ * the result the generator returns.
+ */
+/**
+ * The text of an event stream, returning the stream's final result.
+ *
+ * `for await` reads the text and ignores the return value, which is what
+ * keeps every existing caller working. A caller that wants the tool calls
+ * too can take the return value, or iterate `streamChatEvents()` instead.
+ */
+export declare function streamText(events: AsyncIterable<AIStreamEvent>): AsyncGenerator<string, AIResult, undefined>;
+/**
+ * The JSON objects of a server-sent event stream: each `data:` payload,
+ * parsed. OpenAI's `[DONE]` sentinel ends nothing on its own and is skipped.
+ *
+ * A payload that is not JSON throws. Skipping it, as the drivers used to,
+ * silently drops whatever text or tool-call fragment it carried.
+ */
+export declare function sseJson(body: ReadableStream<Uint8Array>, provider: string): AsyncGenerator<any>;
+/** The JSON objects of a newline-delimited JSON stream, as Ollama sends. */
+export declare function ndjson(body: ReadableStream<Uint8Array>, provider: string): AsyncGenerator<any>;

package/dist/utils/stream.js

@@ -0,0 +1,2 @@
+export async function*streamText(events){let result;for await(const event of events)if(event.type==="text")yield event.text;else if(event.type==="done")result=event.result;if(!result)throw Error("The stream ended without a result");return result}async function*lines(body){const decoder=new TextDecoder;let buffer="";for await(const chunk of body){buffer+=decoder.decode(chunk,{stream:!0});const parts=buffer.split(`
+`);buffer=parts.pop()??"";for(const line of parts)yield line.endsWith("\r")?line.slice(0,-1):line}buffer+=decoder.decode();if(buffer)yield buffer.endsWith("\r")?buffer.slice(0,-1):buffer}export async function*sseJson(body,provider){for await(const line of lines(body)){if(!line.startsWith("data:"))continue;const data=line.slice(5).trimStart();if(data==="[DONE]")continue;yield parseEvent(data,provider)}}export async function*ndjson(body,provider){for await(const line of lines(body))if(line.trim())yield parseEvent(line,provider)}function parseEvent(data,provider){try{return JSON.parse(data)}catch{throw Error(`${provider} sent a stream event that is not JSON: ${data.slice(0,200)}`)}}

package/package.json

@@ -2,7 +2,7 @@
   "name": "@stacksjs/ai",
   "type": "module",
   "sideEffects": false,
-  "version": "0.75.85",
+  "version": "0.75.87",
   "description": "Stacks Artificial Intelligence.",
   "author": "Chris Breuer",
   "contributors": [
@@ -65,10 +65,10 @@
     "prepublishOnly": "bun run build"
   },
   "dependencies": {
-    "@stacksjs/cli": "0.75.85",
-    "@stacksjs/config": "0.75.85",
-    "@stacksjs/storage": "0.75.85",
-    "@stacksjs/ts-cloud": "^0.16.42"
+    "@stacksjs/cli": "0.75.87",
+    "@stacksjs/config": "0.75.87",
+    "@stacksjs/storage": "0.75.87",
+    "@stacksjs/ts-cloud": "^0.16.44"
   },
   "peerDependencies": {
     "@anthropic-ai/claude-agent-sdk": "^0.2.79"