@weiliang0121/capix-model-provider 0.0.0-stage → 0.2.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 wei.liang (https://github.com/weiliang0121)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,3 +1,113 @@
-# Temporary Holding Version
+# 模型 Provider
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+`@weiliang0121/capix-model-provider` 提供 Chat Completions 文本与原生函数调用、可选 SSE，以及唯一的 `ModelReply` 纯值契约。可以从 `.mjs` 使用，不依赖 Runtime、资源、Interaction 或服务。Provider 负责通信和协议接纳；工具选择、参数业务校验、执行、接纳与恢复由调用方组织。
+
+```js
+import { ChatCompletionsProvider, modelReplyText } from '@weiliang0121/capix-model-provider';
+
+const model = new ChatCompletionsProvider({
+  endpoint: 'https://api.deepseek.com/chat/completions',
+  apiKey: process.env.DEEPSEEK_API_KEY,
+  model: process.env.DEEPSEEK_MODEL || 'deepseek-flash',
+  thinking: 'disabled',
+});
+const reply = await model.complete({
+  messages: [{ role: 'user', content: '把 hello 翻译为中文。' }],
+  maxTokens: 512,
+});
+if (reply.termination.reason !== 'stop') throw new Error('模型未正常完成');
+console.log(modelReplyText(reply));
+```
+
+通过 `npm install @weiliang0121/capix-model-provider` 安装后可直接运行；仅在本仓库从源码使用时需要先执行 `pnpm build`。配置由宿主传入，包不读取环境文件。只有显式执行真实请求才会访问模型。
+
+## 唯一回复材料
+
+```ts
+interface ModelReply {
+  formatVersion: 1;
+  id: string;
+  model: string;
+  output: readonly ({ kind: 'text'; text: string } | { kind: 'tool-call'; id: string; name: string; arguments: { encoding: 'json-text'; text: string } })[];
+  termination: {
+    kind: 'completed' | 'incomplete' | 'refused';
+    reason: string;
+  };
+  usage: ModelUsage | null;
+  native: {
+    format: 'chat-completions';
+    version: 1;
+    reasoningContent?: string;
+  };
+}
+```
+
+共同字段保存正文、工具调用、结束原因和用量；`native` 只保存共同字段表达不了的思考文本。`format` 描述协议，不充当厂商身份。Provider 从这份材料编码下一次请求，没有可分别修改的第二份 continuation。工具参数保留原始 JSON 字符串，即使字符串不能被 JSON 解析也不会擅自修复；是否注册、参数是否合法由工具调用方判断。
+
+`stop` 与 `tool_calls` 映射为 `completed`，`content_filter` 为 `refused`，其他原因（包括未知新值）保留原文并映射为 `incomplete`。`completed` 表示协议结束，并不表示业务已完成。`modelReplyText(reply)` 拒绝含工具调用的回复；它不代替业务检查 `stop`、拒绝状态或非空正文。
+
+公开 `requireModelReply`、`requireModelRequest`、`requireToolResult`、`requireToolFeedback` 和 `requireModelExchange` 对未知值作完整校验，返回原对象身份，不克隆、不冻结，也不缓存对可变对象的信任。适用于资源、Interaction、实体与 CP 的接纳边界。校验拒绝旧双字段形状、未知字段/版本、重复 ID、不完整反馈、getter、Proxy、循环引用与非 JSON 值；错误不包含传入正文。Provider 自己生成的回复已深冻结。旧 `text/finishReason` 双字段回复不是新协议的有效持久值，不能自动迁移后冒充已认证反馈。
+
+## 原生工具与续接
+
+```js
+const tools = [
+  {
+    name: 'read_file',
+    description: '读取一个已授权文件。',
+    parameters: {
+      type: 'object',
+      properties: { path: { type: 'string' } },
+      required: ['path'],
+      additionalProperties: false,
+    },
+  },
+];
+const request = { messages: [{ role: 'user', content: '读取 notes.md。' }], tools };
+const first = await model.complete(request);
+
+// 业务负责核验结束状态、参数、权限与执行；此处只展示反馈契约。
+const feedback = first.output
+  .filter((item) => item.kind === 'tool-call')
+  .map((call) => ({
+    toolCallId: call.id,
+    status: 'ok',
+    content: [{ kind: 'text', text: '经授权工具读取的内容' }],
+  }));
+const next = await model.complete({
+  ...request,
+  exchanges: [{ reply: first, feedback }],
+});
+```
+
+`messages` 是固定的 system/user/assistant 文本前缀，`exchanges` 按顺序附加完整的模型交换。每个交换必须有准确且完整的反馈集合；Provider 按模型原调用顺序编码 `assistant.tool_calls` 与 `tool.tool_call_id`，调用方反馈数组的排列不改变关联。`ToolResult` 为 `{status:'ok'|'error',content:[{kind:'text',text}]}`，`ToolFeedback` 另含 `toolCallId`；两种状态都编码为明确的 JSON 数据正文，不能靠措辞暗示失败。工具结果不会成为 system 消息，Provider 也不自动执行其中指令。结果的可信程度、提示策略与实际资源授权仍由宿主和业务确定。
+
+`toolChoice` 可为 `auto`、`none`、`required` 或 `{name}`。指定名称必须已登记；除 `none` 外，显式选择要求有工具目录。参数 Schema 为纯 JSON 对象，原样转交，不在 Provider 内建立另一套业务 Schema 执行器。本实现每份目录/回复最多 32 个工具，工具名支持字母、数字、`_`、`-`，长度 1–128。
+
+DeepSeek 思考模式在带 `tools` 的后续请求中要求回传此前回复的 `reasoning_content`，包括没有工具调用的回复。Provider 在 exchanges 中始终保留并编码该字段；不带 tools 时，是否使用由兼容端点决定。调用方不能用只有正文的普通 assistant 消息替代需要精确续接的思考回复。协议依据：[工具调用](https://api-docs.deepseek.com/guides/tool_calls/)、[思考模式](https://api-docs.deepseek.com/guides/thinking_mode/)。
+
+## 流式与用量
+
+```js
+const reply = await model.complete({ messages: [{ role: 'user', content: '解释这份材料。' }], stream: true }, { signal, emit: (delta) => process.stdout.write(delta) });
+```
+
+`stream` 默认 false；单独提供 emit 不开启流式。emit 只同步交付正文预览，工具参数与思考文本不会被当作正文发布。未提供 emit 时仍完整消费。回调抛错保留原错误身份，回调返回 Promise 会被拒绝。预览不代表最终回复或业务接纳已成立。
+
+SSE 按 `tool_calls[index]` 拼接名称与参数片段，保留 reasoning_content，终态与独立 usage 均支持。缺少 ID、重复 ID、索引缺口、超过 32 个调用、终态后继续 choice、重复 usage、缺少 finish_reason 或 `[DONE]` 均拒绝。`[DONE]` 结束协议后立即释放读取，不要求上游 HTTP 连接先关闭；不重发或退回非流式请求。原有 UTF-8 分片、SSE 换行/注释、多行 data、接收者失败、取消和总超时行为保持。
+
+JSON 与 SSE 共用 `ModelUsage`：`promptTokens / completionTokens / totalTokens`，可选 `promptCacheHitTokens / promptCacheMissTokens` 原样保留。计数必须是非负安全整数，缓存单项不超过 promptTokens，两项都报告时其和须等于 promptTokens。缺失不估算为零；不把其他厂商的缓存字段自动映射成 DeepSeek 计数。
+
+## 容量与范围
+
+`ChatCompletionsProvider.inspectRequest(request)` 同步完成请求预检，返回冻结的 `{requestBytes, wireBytes}`：前者是共同请求完整 JSON 的 UTF-8 字节数，后者是实际 HTTP 正文字节数，包括固定的 model/thinking 配置、工具目录、交换编码和流式字段。它与 `complete` 共用校验和编码路径，能够计算 Unicode、转义字符以及工具反馈嵌套 JSON 带来的真实大小；不发请求、不读取网络，也不返回正文、headers 或密钥。
+
+`ModelProvider` 上此端口可选，依赖精确发送预算的宿主应在装配时要求它存在。无效或超出共同请求／HTTP 正文 2 MiB 限额时，预检与 `complete` 一样抛出 `ProviderError('REQUEST')`。预检不冻结输入、不预留额度，也不保证模型上下文 token 容量；实际发送仍重新校验当前请求，修改过输入后必须以本次检查为准。
+
+- 公开请求、回复、反馈、交换值均限制为 2 MiB 的纯 JSON，深度最多 64、值数最多 100,000。请求编码后的 HTTP 正文也受 2 MiB 限制。合法的单项不代表任意累计历史都能继续装入请求。
+- 默认总超时 60 秒，涵盖发送与完整读取。`maxResponseBytes` 是独立的实际传输预算，默认 2 MiB，SSE 计入事件封装、注释和思考内容；提高传输预算不会放宽公开回复的纯值上限。
+- 一次 complete 只有一次 HTTP 请求，无隐式重试。调用取消保持原 reason，不修改调用者 signal；对独立 Interaction 的生命周期不作推断。
+- `ProviderError` 只含 code/message/可选 HTTP status，不保存上游正文、请求、headers、密钥或底层错误。宿主可注入 fetch；实现必须遵守 Fetch 取消与重定向语义。
+- `thinking` 是宿主显式配置的 DeepSeek 扩展。当前实现 Chat Completions 函数调用与文本；不承诺 Responses、Anthropic 原生协议、MCP、多模态或 strict beta。正式 Transcript 的消息角色不因此扩展。
+
+当前架构与限制见[原生工具调用](../../docs/native-tool-integration.md)。完整协议参考 [Chat Completions](https://api-docs.deepseek.com/api/create-chat-completion/) 与 [SSE 标准](https://html.spec.whatwg.org/multipage/server-sent-events.html#event-stream-interpretation)。

package/dist/contract.d.ts

@@ -0,0 +1,9 @@
+import type { ModelExchange, ModelReply, ModelRequest, ToolFeedback, ToolResult } from './types.js';
+export declare function terminationKind(reason: string): ModelReply['termination']['kind'];
+export declare function requireModelReply(value: unknown): ModelReply;
+export declare function requireToolResult(value: unknown): ToolResult;
+export declare function requireToolFeedback(value: unknown): ToolFeedback;
+export declare function requireModelExchange(value: unknown): ModelExchange;
+export declare function requireModelRequest(value: unknown): ModelRequest;
+export declare function measureModelRequest(value: unknown): number;
+export declare function modelReplyText(reply: ModelReply): string;

package/dist/contract.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{count,identity,record,requireJson,shape,toolLimit}from"./value.js";function terminationKind(reason){return reason==="stop"||reason==="tool_calls"?"completed":reason==="content_filter"?"refused":"incomplete"}__name(terminationKind,"terminationKind");function requireModelReply(value){return requireJson(value),checkReply(value),value}__name(requireModelReply,"requireModelReply");function checkReply(value){if(shape(value,["formatVersion","id","model","output","termination","usage","native"]),value.formatVersion!==1)throw new TypeError("Unsupported model reply version");if(identity(value.id),identity(value.model),shape(value.termination,["kind","reason"]),identity(value.termination.reason),value.termination.kind!==terminationKind(value.termination.reason))throw new TypeError("Model termination does not match its reason");if(shape(value.native,["format","version"],["reasoningContent"]),value.native.format!=="chat-completions"||value.native.version!==1||"reasoningContent"in value.native&&typeof value.native.reasoningContent!="string")throw new TypeError("Unsupported model native material");if(!Array.isArray(value.output)||value.output.length>toolLimit+1)throw new TypeError("Model output is invalid");let text=!1;const calls=new Set;for(const item of value.output)if(record(item)&&item.kind==="text"){if(checkText(item),text||calls.size)throw new TypeError("Unsupported Chat Completions output order");text=!0}else{if(shape(item,["kind","id","name","arguments"]),item.kind!=="tool-call")throw new TypeError("Unsupported model output kind");if(identity(item.id),toolName(item.name),shape(item.arguments,["encoding","text"]),item.arguments.encoding!=="json-text"||typeof item.arguments.text!="string")throw new TypeError("Unsupported tool argument representation");if(calls.has(item.id)||calls.size>=toolLimit)throw new TypeError("Duplicate or excessive model tool calls");calls.add(item.id)}if(value.termination.reason==="tool_calls"&&!calls.size||value.termination.reason==="stop"&&calls.size)throw new TypeError("Model termination disagrees with tool output");if(value.usage!==null){shape(value.usage,["promptTokens","completionTokens","totalTokens"],["promptCacheHitTokens","promptCacheMissTokens"]);const{promptTokens,completionTokens,totalTokens,promptCacheHitTokens:hit,promptCacheMissTokens:miss}=value.usage;if(!count(promptTokens)||!count(completionTokens)||!count(totalTokens)||hit!==void 0&&(!count(hit)||hit>promptTokens)||miss!==void 0&&(!count(miss)||miss>promptTokens)||typeof hit=="number"&&typeof miss=="number"&&hit!==promptTokens-miss)throw new TypeError("Invalid model usage counts")}}__name(checkReply,"checkReply");function checkText(value){if(shape(value,["kind","text"]),value.kind!=="text"||typeof value.text!="string")throw new TypeError("Tool content must contain text blocks")}__name(checkText,"checkText");function toolName(value){if(typeof value!="string"||!/^[A-Za-z0-9_-]{1,128}$/u.test(value))throw new TypeError("Invalid tool name")}__name(toolName,"toolName");function checkResult(value,feedback){if(shape(value,feedback?["toolCallId","status","content"]:["status","content"]),feedback&&identity(value.toolCallId),typeof value.status!="string"||!["ok","error"].includes(value.status)||!Array.isArray(value.content)||!value.content.length||value.content.length>32)throw new TypeError("Invalid tool feedback content");for(const item of value.content)checkText(item)}__name(checkResult,"checkResult");function requireToolResult(value){return requireJson(value),checkResult(value,!1),value}__name(requireToolResult,"requireToolResult");function requireToolFeedback(value){return requireJson(value),checkResult(value,!0),value}__name(requireToolFeedback,"requireToolFeedback");function checkExchange(value){if(shape(value,["reply","feedback"]),checkReply(value.reply),value.reply.termination.kind!=="completed")throw new TypeError("Incomplete or refused reply cannot be continued as a closed exchange");if(!Array.isArray(value.feedback))throw new TypeError("Tool feedback must be an array");const calls=value.reply.output.filter(item=>item.kind==="tool-call");if(value.feedback.length!==calls.length)throw new TypeError("Tool feedback does not close the model exchange");const ids=new Set;for(const feedback of value.feedback){checkResult(feedback,!0);const id=feedback.toolCallId;if(ids.has(id)||!calls.some(call=>call.id===id))throw new TypeError("Tool feedback has a duplicate or foreign call identity");ids.add(id)}}__name(checkExchange,"checkExchange");function requireModelExchange(value){return requireJson(value),checkExchange(value),value}__name(requireModelExchange,"requireModelExchange");function requireModelRequest(value){return measureModelRequest(value),value}__name(requireModelRequest,"requireModelRequest");function measureModelRequest(value){const bytes=requireJson(value);return checkRequest(value),bytes}__name(measureModelRequest,"measureModelRequest");function checkRequest(value){if(shape(value,["messages"],["exchanges","tools","toolChoice","maxTokens","responseFormat","temperature","stream"]),!Array.isArray(value.messages)||!value.messages.length)throw new TypeError("At least one model message is required");for(const message of value.messages)if(shape(message,["role","content"]),typeof message.role!="string"||!["system","user","assistant"].includes(message.role)||typeof message.content!="string")throw new TypeError("Model message has an unsupported role or content");if(value.exchanges!==void 0){if(!Array.isArray(value.exchanges))throw new TypeError("Model exchanges must be an array");for(const exchange of value.exchanges)checkExchange(exchange)}const names=new Set;if(value.tools!==void 0){if(!Array.isArray(value.tools)||!value.tools.length||value.tools.length>toolLimit)throw new TypeError("Model tool catalog must contain 1 to 32 tools");for(const tool of value.tools){if(shape(tool,["name","description","parameters"]),toolName(tool.name),names.has(tool.name)||typeof tool.description!="string"||!record(tool.parameters))throw new TypeError("Invalid or duplicate model tool declaration");names.add(tool.name)}}if(value.toolChoice!==void 0){if(record(value.toolChoice)){if(shape(value.toolChoice,["name"]),toolName(value.toolChoice.name),!names.has(value.toolChoice.name))throw new TypeError("Selected model tool is not declared")}else if(typeof value.toolChoice!="string"||!["auto","none","required"].includes(value.toolChoice))throw new TypeError("Unsupported model tool choice");if(value.toolChoice!=="none"&&!names.size)throw new TypeError("Model tool choice requires tools")}if(value.maxTokens!==void 0&&(!count(value.maxTokens)||value.maxTokens<1))throw new TypeError("maxTokens must be a positive integer");if(value.temperature!==void 0&&(typeof value.temperature!="number"||!Number.isFinite(value.temperature)||value.temperature<0||value.temperature>2))throw new TypeError("temperature must be between zero and two");if(value.responseFormat!==void 0&&(typeof value.responseFormat!="string"||!["text","json_object"].includes(value.responseFormat)))throw new TypeError("Unsupported response format");if(value.stream!==void 0&&typeof value.stream!="boolean")throw new TypeError("stream must be a boolean")}__name(checkRequest,"checkRequest");function modelReplyText(reply){let text="";for(const item of reply.output){if(item.kind==="tool-call")throw new TypeError("Tool-call output cannot be projected as a text reply");text+=item.text}return text}__name(modelReplyText,"modelReplyText");export{measureModelRequest,modelReplyText,requireModelExchange,requireModelReply,requireModelRequest,requireToolFeedback,requireToolResult,terminationKind};

package/dist/error.d.ts

@@ -0,0 +1,10 @@
+export declare class ProviderError extends Error {
+    readonly code: 'CONFIGURATION' | 'REQUEST' | 'NETWORK' | 'HTTP' | 'TIMEOUT' | 'RESPONSE_LIMIT' | 'ENVELOPE';
+    readonly status?: number | undefined;
+    constructor(code: 'CONFIGURATION' | 'REQUEST' | 'NETWORK' | 'HTTP' | 'TIMEOUT' | 'RESPONSE_LIMIT' | 'ENVELOPE', message: string, status?: number | undefined);
+}
+export declare class ModelEmissionFailure {
+    readonly failure: unknown;
+    constructor(failure: unknown);
+}
+export declare function isRequestCancellation(error: unknown, signal: AbortSignal): boolean;

package/dist/error.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});class ProviderError extends Error{static{__name(this,"ProviderError")}code;status;constructor(code,message,status){super(message),this.code=code,this.status=status,this.name="ProviderError"}}class ModelEmissionFailure{static{__name(this,"ModelEmissionFailure")}failure;constructor(failure){this.failure=failure}}function isRequestCancellation(error,signal){return signal.aborted&&(error===signal.reason||error instanceof DOMException&&error.name==="AbortError")}__name(isRequestCancellation,"isRequestCancellation");export{ModelEmissionFailure,ProviderError,isRequestCancellation};

package/dist/index.d.ts

@@ -0,0 +1,4 @@
+export { ChatCompletionsProvider } from './provider.js';
+export { ProviderError } from './error.js';
+export { requireModelReply, requireModelRequest, requireModelExchange, requireToolResult, requireToolFeedback, modelReplyText } from './contract.js';
+export type { ChatCompletionsConfiguration, ModelControls, ModelMessage, ModelProvider, ModelReply, ModelRequest, ModelRequestInspection, ModelOutput, ModelTextOutput, ModelToolCall, ModelUsage, ModelExchange, ModelTool, ModelToolChoice, ToolResult, ToolFeedback, JsonObject, JsonValue } from './types.js';

package/dist/index.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+import{ChatCompletionsProvider}from"./provider.js";import{ProviderError}from"./error.js";import{requireModelReply,requireModelRequest,requireModelExchange,requireToolResult,requireToolFeedback,modelReplyText}from"./contract.js";export{ChatCompletionsProvider,ProviderError,modelReplyText,requireModelExchange,requireModelReply,requireModelRequest,requireToolFeedback,requireToolResult};

package/dist/provider.d.ts

@@ -0,0 +1,9 @@
+import type { ChatCompletionsConfiguration, ModelControls, ModelProvider, ModelReply, ModelRequest, ModelRequestInspection } from './types.js';
+export declare class ChatCompletionsProvider implements ModelProvider {
+    #private;
+    constructor(configuration: ChatCompletionsConfiguration);
+    inspectRequest(request: ModelRequest): ModelRequestInspection;
+    complete(request: ModelRequest, controls?: ModelControls): Promise<ModelReply>;
+    private prepareRequest;
+    private requestBody;
+}

package/dist/provider.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{isRequestCancellation,ModelEmissionFailure,ProviderError}from"./error.js";import{readModelReply}from"./response.js";import{readModelStream}from"./stream.js";import{measureModelRequest}from"./contract.js";import{encodeMessages}from"./wire.js";import{valueLimit}from"./value.js";class ChatCompletionsProvider{static{__name(this,"ChatCompletionsProvider")}#configuration;#fetch;constructor(configuration){let endpoint;try{endpoint=new URL(configuration.endpoint)}catch{throw new ProviderError("CONFIGURATION","A complete model HTTP endpoint is required")}if(!["http:","https:"].includes(endpoint.protocol)||endpoint.username||endpoint.password||endpoint.hash)throw new ProviderError("CONFIGURATION","A plain HTTP endpoint without embedded credentials or fragment is required");if(typeof configuration.apiKey!="string"||!configuration.apiKey.trim()||/[\r\n]/.test(configuration.apiKey))throw new ProviderError("CONFIGURATION","A model API key is required");if(typeof configuration.model!="string"||!configuration.model.trim())throw new ProviderError("CONFIGURATION","A model name is required");const timeoutMs=configuration.timeoutMs??6e4,maxResponseBytes=configuration.maxResponseBytes??2*1024*1024;if(!Number.isSafeInteger(timeoutMs)||timeoutMs<1||timeoutMs>2147483647||!Number.isSafeInteger(maxResponseBytes)||maxResponseBytes<1)throw new ProviderError("CONFIGURATION","Positive bounded timeout and response byte limit are required");if(configuration.thinking!==void 0&&!["enabled","disabled"].includes(configuration.thinking))throw new ProviderError("CONFIGURATION","Unsupported thinking setting");if(configuration.fetch!==void 0&&typeof configuration.fetch!="function")throw new ProviderError("CONFIGURATION","Fetch transport must be a function");this.#configuration=Object.freeze({endpoint:endpoint.href,apiKey:configuration.apiKey,model:configuration.model,timeoutMs,maxResponseBytes,thinking:configuration.thinking}),this.#fetch=configuration.fetch??globalThis.fetch}inspectRequest(request){return this.prepareRequest(request).inspection}async complete(request,controls={}){const external=controls.signal;external?.throwIfAborted();const emit=controls.emit;if(emit!==void 0&&typeof emit!="function")throw new ProviderError("REQUEST","Model preview consumer must be a function");const{body,streaming}=this.prepareRequest(request),controller=new AbortController,abort=__name(()=>controller.abort(external?.reason),"abort");external?.addEventListener("abort",abort,{once:!0}),external?.aborted&&abort();const timeout=setTimeout(()=>controller.abort(new ProviderError("TIMEOUT","Model request exceeded its total timeout")),this.#configuration.timeoutMs);try{controller.signal.throwIfAborted();const response=await this.#fetch(this.#configuration.endpoint,{method:"POST",redirect:"error",headers:{"Content-Type":"application/json",Authorization:`Bearer ${this.#configuration.apiKey}`},body,signal:controller.signal});if(controller.signal.aborted&&(await response.body?.cancel().catch(()=>{}),controller.signal.throwIfAborted()),!response.ok){const failure=new ProviderError("HTTP",`Model endpoint returned HTTP ${response.status}; request was not retried`,response.status);throw await response.body?.cancel().catch(()=>{}),failure}return streaming?await readModelStream(response,controller.signal,this.#configuration.maxResponseBytes,emit):await readModelReply(response,controller.signal,this.#configuration.maxResponseBytes)}catch(error){throw error instanceof ModelEmissionFailure?error.failure:isRequestCancellation(error,controller.signal)?controller.signal.reason:error instanceof ProviderError?error:new ProviderError("NETWORK","Model transport failed; request was not retried")}finally{clearTimeout(timeout),external?.removeEventListener("abort",abort)}}prepareRequest(request){try{const requestBytes=measureModelRequest(request),streaming=request.stream===!0,body=this.requestBody(request,streaming),wireBytes=Buffer.byteLength(body);if(wireBytes>valueLimit)throw new TypeError("Model wire request exceeds byte limit");return{body,streaming,inspection:Object.freeze({requestBytes,wireBytes})}}catch{throw new ProviderError("REQUEST","Model request has invalid, unsupported or excessive content")}}requestBody(request,stream){const messages=encodeMessages(request),tools=request.tools?.map(tool=>({type:"function",function:{name:tool.name,description:tool.description,parameters:tool.parameters}})),choice=request.toolChoice;return JSON.stringify({model:this.#configuration.model,messages,stream,...tools?{tools}:{},...choice===void 0?{}:{tool_choice:typeof choice=="string"?choice:{type:"function",function:{name:choice.name}}},...stream?{stream_options:{include_usage:!0}}:{},...request.maxTokens===void 0?{}:{max_tokens:request.maxTokens},...request.temperature===void 0?{}:{temperature:request.temperature},...request.responseFormat===void 0?{}:{response_format:{type:request.responseFormat}},...this.#configuration.thinking===void 0?{}:{thinking:{type:this.#configuration.thinking}}})}}export{ChatCompletionsProvider};

package/dist/response.d.ts

@@ -0,0 +1,2 @@
+import type { ModelReply } from './types.js';
+export declare function readModelReply(response: Response, signal: AbortSignal, limit: number): Promise<ModelReply>;

package/dist/response.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{isRequestCancellation,ProviderError}from"./error.js";import{acceptWireReply}from"./wire.js";async function readModelReply(response,signal,limit){if(!response.body)throw envelope();const reader=response.body.getReader(),abort=__name(()=>{reader.cancel(signal.reason).catch(()=>{})},"abort");signal.addEventListener("abort",abort,{once:!0});let complete=!1;try{signal.throwIfAborted();const decoder=new TextDecoder("utf-8",{fatal:!0});let bytes=0,text="",chunk=await reader.read();for(;!chunk.done;){if(signal.throwIfAborted(),bytes+=chunk.value.byteLength,bytes>limit)throw new ProviderError("RESPONSE_LIMIT","Model response exceeds the configured byte limit");text+=decoder.decode(chunk.value,{stream:!0}),chunk=await reader.read()}return signal.throwIfAborted(),text+=decoder.decode(),complete=!0,acceptWireReply(JSON.parse(text))}catch(error){throw isRequestCancellation(error,signal)?signal.reason:error instanceof ProviderError?error:envelope()}finally{signal.removeEventListener("abort",abort),complete||await reader.cancel().catch(()=>{}),reader.releaseLock()}}__name(readModelReply,"readModelReply");function envelope(){return new ProviderError("ENVELOPE","Model response is not a supported Chat Completions envelope")}__name(envelope,"envelope");export{readModelReply};

package/dist/stream.d.ts

@@ -0,0 +1,2 @@
+import type { ModelReply } from './types.js';
+export declare function readModelStream(response: Response, signal: AbortSignal, limit: number, emit?: (delta: string) => void): Promise<ModelReply>;

package/dist/stream.js

@@ -0,0 +1,5 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{isRequestCancellation,ModelEmissionFailure,ProviderError}from"./error.js";import{acceptUsage}from"./usage.js";import{acceptWireReply}from"./wire.js";import{toolLimit}from"./value.js";async function readModelStream(response,signal,limit,emit){if(!response.body)throw envelope();if(response.headers.get("content-type")?.split(";")[0]?.trim().toLowerCase()!=="text/event-stream")throw await response.body.cancel().catch(()=>{}),envelope();const reader=response.body.getReader(),abort=__name(()=>{reader.cancel(signal.reason).catch(()=>{})},"abort");signal.addEventListener("abort",abort,{once:!0});try{signal.throwIfAborted();const completion=new ChatCompletionStream(signal,emit),events=new ServerSentEvents(data=>completion.receive(data)),decoder=new TextDecoder("utf-8",{fatal:!0});let bytes=0,chunk=await reader.read();for(;!chunk.done&&!completion.done;){if(signal.throwIfAborted(),bytes+=chunk.value.byteLength,bytes>limit)throw new ProviderError("RESPONSE_LIMIT","Model response exceeds the configured byte limit");events.write(decoder.decode(chunk.value,{stream:!0})),completion.done||(chunk=await reader.read())}if(signal.throwIfAborted(),!completion.done)throw events.write(decoder.decode()),envelope();return completion.result()}catch(error){throw error instanceof ModelEmissionFailure?error:isRequestCancellation(error,signal)?signal.reason:error instanceof ProviderError?error:envelope()}finally{signal.removeEventListener("abort",abort),await reader.cancel().catch(()=>{}),reader.releaseLock()}}__name(readModelStream,"readModelStream");class ServerSentEvents{static{__name(this,"ServerSentEvents")}receive;#line="";#skipLF=!1;#data=[];#done=!1;constructor(receive){this.receive=receive}write(text){let start=0;for(let index=0;index<text.length&&!this.#done;index++){const character=text[index];if(this.#skipLF&&(this.#skipLF=!1,character===`
+`)){start=index+1;continue}character!==`
+`&&character!=="\r"||(this.acceptLine(this.#line+text.slice(start,index)),this.#line="",this.#skipLF=character==="\r",start=index+1)}this.#done||(this.#line+=text.slice(start))}acceptLine(line){if(line===""){this.#data.length&&(this.#done=this.receive(this.#data.join(`
+`))),this.#data=[];return}const colon=line.indexOf(":");if((colon===-1?line:line.slice(0,colon))!=="data")return;const value=colon===-1?"":line.slice(colon+1);this.#data.push(value.startsWith(" ")?value.slice(1):value)}}class ChatCompletionStream{static{__name(this,"ChatCompletionStream")}signal;emit;#id;#model;#finishReason;#usage=null;#wireUsage;#fragments=[];#hasContent=!1;#reasoning=[];#hasReasoning=!1;#calls=new Map;#done=!1;constructor(signal,emit){this.signal=signal,this.emit=emit}get done(){return this.#done}receive(data){if(this.signal.throwIfAborted(),data==="[DONE]"){if(!this.#id||!this.#model||!this.#finishReason)throw envelope();return this.#done=!0,!0}const value=JSON.parse(data);if(!record(value)||typeof value.id!="string"||!value.id||typeof value.model!="string"||!value.model||!Array.isArray(value.choices)||this.#id!==void 0&&this.#id!==value.id||this.#model!==void 0&&this.#model!==value.model||value.error!=null)throw envelope();this.#id=value.id,this.#model=value.model;const usage=acceptUsage(value.usage);if(usage&&this.#usage)throw envelope();if(value.choices.length===0){if(!this.#finishReason||!usage)throw envelope();return this.#usage=usage,this.#wireUsage=value.usage,!1}if(this.#finishReason||value.choices.length!==1)throw envelope();const choice=value.choices[0];if(!record(choice)||choice.index!==0||!record(choice.delta))throw envelope();const delta=choice.delta;if(delta.role!==void 0&&delta.role!=="assistant"||delta.function_call!=null||delta.refusal!=null||choice.message!=null||delta.reasoning_content!=null&&typeof delta.reasoning_content!="string"||delta.content!=null&&typeof delta.content!="string")throw envelope();const finishReason=choice.finish_reason;if(finishReason!=null&&(typeof finishReason!="string"||!finishReason)||usage&&!finishReason)throw envelope();if(typeof finishReason=="string"&&(this.#finishReason=finishReason),usage&&(this.#usage=usage,this.#wireUsage=value.usage),typeof delta.reasoning_content=="string"&&(this.#reasoning.push(delta.reasoning_content),this.#hasReasoning=!0),delta.tool_calls!=null){if(!Array.isArray(delta.tool_calls)||delta.tool_calls.length>toolLimit)throw envelope();for(const fragment of delta.tool_calls){if(!record(fragment)||typeof fragment.index!="number"||!Number.isSafeInteger(fragment.index)||fragment.index<0||fragment.index>=toolLimit||fragment.type!==void 0&&fragment.type!=="function"||fragment.id!==void 0&&(typeof fragment.id!="string"||!fragment.id)||fragment.function!==void 0&&!record(fragment.function))throw envelope();let call=this.#calls.get(fragment.index);if(call||(call={id:"",type:"function",function:{name:"",arguments:""}},this.#calls.set(fragment.index,call)),typeof fragment.id=="string"){if(call.id)throw envelope();call.id=fragment.id}if(record(fragment.function)){const{name,arguments:args}=fragment.function;if(name!==void 0&&typeof name!="string"||args!==void 0&&typeof args!="string")throw envelope();typeof name=="string"&&(call.function.name+=name),typeof args=="string"&&(call.function.arguments+=args)}}}if(typeof delta.content=="string"&&(this.#hasContent=!0),typeof delta.content=="string"&&delta.content.length){if(this.#fragments.push(delta.content),this.emit)try{const returned=this.emit(delta.content);if((typeof returned=="object"&&returned!==null||typeof returned=="function")&&"then"in returned&&typeof returned.then=="function")throw Promise.resolve(returned).catch(()=>{}),new TypeError("Model preview consumer must be synchronous")}catch(failure){throw new ModelEmissionFailure(failure)}this.signal.throwIfAborted()}return!1}result(){if(!this.#done||!this.#id||!this.#model||!this.#finishReason)throw envelope();const indexes=[...this.#calls.keys()].sort((left,right)=>left-right);if(indexes.some((index,position)=>index!==position))throw envelope();return acceptWireReply({id:this.#id,model:this.#model,choices:[{index:0,message:{role:"assistant",content:this.#hasContent||!this.#calls.size?this.#fragments.join(""):null,...this.#hasReasoning?{reasoning_content:this.#reasoning.join("")}:{},...this.#calls.size?{tool_calls:indexes.map(index=>this.#calls.get(index))}:{}},finish_reason:this.#finishReason}],usage:this.#wireUsage})}}function record(value){return typeof value=="object"&&value!==null&&!Array.isArray(value)}__name(record,"record");function envelope(){return new ProviderError("ENVELOPE","Model response is not a supported Chat Completions stream")}__name(envelope,"envelope");export{readModelStream};

package/dist/types.d.ts

@@ -0,0 +1,95 @@
+export type JsonValue = null | boolean | number | string | readonly JsonValue[] | JsonObject;
+export interface JsonObject {
+    readonly [key: string]: JsonValue;
+}
+export interface ModelMessage {
+    readonly role: 'system' | 'user' | 'assistant';
+    readonly content: string;
+}
+export interface ModelTextOutput {
+    readonly kind: 'text';
+    readonly text: string;
+}
+export interface ModelToolCall {
+    readonly kind: 'tool-call';
+    readonly id: string;
+    readonly name: string;
+    readonly arguments: {
+        readonly encoding: 'json-text';
+        readonly text: string;
+    };
+}
+export type ModelOutput = ModelTextOutput | ModelToolCall;
+export interface ModelUsage {
+    readonly promptTokens: number;
+    readonly completionTokens: number;
+    readonly totalTokens: number;
+    readonly promptCacheHitTokens?: number;
+    readonly promptCacheMissTokens?: number;
+}
+export interface ModelReply {
+    readonly formatVersion: 1;
+    readonly id: string;
+    readonly model: string;
+    readonly output: readonly ModelOutput[];
+    readonly termination: {
+        readonly kind: 'completed' | 'incomplete' | 'refused';
+        readonly reason: string;
+    };
+    readonly usage: ModelUsage | null;
+    readonly native: {
+        readonly format: 'chat-completions';
+        readonly version: 1;
+        readonly reasoningContent?: string;
+    };
+}
+export interface ToolResult {
+    readonly status: 'ok' | 'error';
+    readonly content: readonly ModelTextOutput[];
+}
+export interface ToolFeedback extends ToolResult {
+    readonly toolCallId: string;
+}
+export interface ModelExchange {
+    readonly reply: ModelReply;
+    readonly feedback: readonly ToolFeedback[];
+}
+export interface ModelTool {
+    readonly name: string;
+    readonly description: string;
+    readonly parameters: JsonObject;
+}
+export type ModelToolChoice = 'auto' | 'none' | 'required' | {
+    readonly name: string;
+};
+export interface ModelRequest {
+    readonly messages: readonly ModelMessage[];
+    readonly exchanges?: readonly ModelExchange[];
+    readonly tools?: readonly ModelTool[];
+    readonly toolChoice?: ModelToolChoice;
+    readonly maxTokens?: number;
+    readonly responseFormat?: 'text' | 'json_object';
+    readonly temperature?: number;
+    readonly stream?: boolean;
+}
+export interface ModelControls {
+    readonly signal?: AbortSignal;
+    readonly emit?: (delta: string) => void;
+}
+export interface ModelProvider {
+    inspectRequest?(request: ModelRequest): ModelRequestInspection;
+    complete(request: ModelRequest, controls?: ModelControls): Promise<ModelReply>;
+}
+export interface ModelRequestInspection {
+    readonly requestBytes: number;
+    readonly wireBytes: number;
+}
+export interface ChatCompletionsConfiguration {
+    readonly endpoint: string;
+    readonly apiKey: string;
+    readonly model: string;
+    readonly timeoutMs?: number;
+    readonly maxResponseBytes?: number;
+    readonly thinking?: 'enabled' | 'disabled';
+    readonly fetch?: typeof fetch;
+}

package/dist/types.js

@@ -0,0 +1 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */

package/dist/usage.d.ts

@@ -0,0 +1,2 @@
+import type { ModelReply } from './types.js';
+export declare function acceptUsage(value: unknown): ModelReply['usage'];

package/dist/usage.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{ProviderError}from"./error.js";function acceptUsage(value){if(value==null)return null;if(typeof value!="object"||Array.isArray(value)||!("prompt_tokens"in value)||!("completion_tokens"in value)||!("total_tokens"in value)||!count(value.prompt_tokens)||!count(value.completion_tokens)||!count(value.total_tokens))throw invalidUsage();const hit="prompt_cache_hit_tokens"in value?value.prompt_cache_hit_tokens:void 0,miss="prompt_cache_miss_tokens"in value?value.prompt_cache_miss_tokens:void 0;if(hit!==void 0&&(!count(hit)||hit>value.prompt_tokens)||miss!==void 0&&(!count(miss)||miss>value.prompt_tokens)||typeof hit=="number"&&typeof miss=="number"&&hit!==value.prompt_tokens-miss)throw invalidUsage();return Object.freeze({promptTokens:value.prompt_tokens,completionTokens:value.completion_tokens,totalTokens:value.total_tokens,...hit===void 0?{}:{promptCacheHitTokens:hit},...miss===void 0?{}:{promptCacheMissTokens:miss}})}__name(acceptUsage,"acceptUsage");function count(value){return typeof value=="number"&&Number.isSafeInteger(value)&&value>=0}__name(count,"count");function invalidUsage(){return new ProviderError("ENVELOPE","Model response contains invalid usage counts")}__name(invalidUsage,"invalidUsage");export{acceptUsage};

package/dist/value.d.ts

@@ -0,0 +1,8 @@
+export declare const valueLimit: number;
+export declare const toolLimit = 32;
+export declare function requireJson(value: unknown): number;
+export declare function freezeJson<T>(value: T): T;
+export declare function record(value: unknown): value is Record<string, unknown>;
+export declare function shape(value: unknown, required: readonly string[], optional?: readonly string[]): asserts value is Record<string, unknown>;
+export declare function identity(value: unknown): asserts value is string;
+export declare function count(value: unknown): value is number;

package/dist/value.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{isProxy}from"node:util/types";const valueLimit=2*1024*1024,toolLimit=32;function requireJson(value){let nodes=0,bytes=0;const ancestors=new Set,add=__name(size=>{if(bytes+=size,bytes>valueLimit)throw new TypeError("Model value exceeds byte limit")},"add"),visit=__name((item,depth)=>{if(++nodes>1e5||depth>64)throw new TypeError("Model value exceeds structural limits");if(item===null||typeof item=="boolean"){add(item===null||item?4:5);return}if(typeof item=="string"){if(Buffer.byteLength(item)>valueLimit)throw new TypeError("Model value exceeds byte limit");add(Buffer.byteLength(JSON.stringify(item)));return}if(typeof item=="number"&&Number.isFinite(item)){add(Buffer.byteLength(JSON.stringify(item)));return}if(typeof item!="object"||isProxy(item)||ancestors.has(item))throw new TypeError("Model value must be acyclic JSON");const array=Array.isArray(item);if(!array&&Object.getPrototypeOf(item)!==Object.prototype&&Object.getPrototypeOf(item)!==null)throw new TypeError("Model value must contain plain objects");const descriptors=Object.getOwnPropertyDescriptors(item),keys=Reflect.ownKeys(descriptors);if(array&&keys.length!==item.length+1)throw new TypeError("Model arrays must be dense");const length=keys.length-(array?1:0);add(2+Math.max(0,length-1)),ancestors.add(item);for(const key of keys){if(array&&key==="length")continue;if(typeof key!="string")throw new TypeError("Model value cannot contain symbol fields");const descriptor=descriptors[key];if(!descriptor.enumerable||!("value"in descriptor))throw new TypeError("Model value cannot contain accessors or hidden fields");if(array&&(!/^(0|[1-9][0-9]*)$/u.test(key)||Number(key)>=item.length))throw new TypeError("Model arrays cannot have extra fields");array||add(Buffer.byteLength(JSON.stringify(key))+1),visit(descriptor.value,depth+1)}ancestors.delete(item)},"visit");return visit(value,0),bytes}__name(requireJson,"requireJson");function freezeJson(value){if(value!==null&&typeof value=="object"){for(const child of Object.values(value))freezeJson(child);Object.freeze(value)}return value}__name(freezeJson,"freezeJson");function record(value){return value!==null&&typeof value=="object"&&!Array.isArray(value)}__name(record,"record");function shape(value,required,optional=[]){if(!record(value)||required.some(key=>!Object.hasOwn(value,key))||Object.keys(value).some(key=>!required.includes(key)&&!optional.includes(key)))throw new TypeError("Model value has unsupported or missing fields")}__name(shape,"shape");function identity(value){if(typeof value!="string"||value.length<1||value.length>256)throw new TypeError("Model identity is invalid")}__name(identity,"identity");function count(value){return typeof value=="number"&&Number.isSafeInteger(value)&&value>=0}__name(count,"count");export{count,freezeJson,identity,record,requireJson,shape,toolLimit,valueLimit};

package/dist/wire.d.ts

@@ -0,0 +1,3 @@
+import type { ModelReply, ModelRequest } from './types.js';
+export declare function acceptWireReply(value: unknown): ModelReply;
+export declare function encodeMessages(request: ModelRequest): unknown[];

package/dist/wire.js

@@ -0,0 +1,2 @@
+/*! MIT License | Copyright (c) 2026 wei.liang | See LICENSE */
+var __defProp=Object.defineProperty;var __name=(target,value)=>__defProp(target,"name",{value,configurable:!0});import{requireModelReply,terminationKind}from"./contract.js";import{ProviderError}from"./error.js";import{acceptUsage}from"./usage.js";import{freezeJson,record,toolLimit}from"./value.js";function acceptWireReply(value){if(!record(value)||!Array.isArray(value.choices)||value.choices.length!==1||value.error!=null)throw envelope();const choice=value.choices[0];if(!record(choice)||choice.index!==void 0&&choice.index!==0||!record(choice.message)||typeof choice.finish_reason!="string"||!choice.finish_reason)throw envelope();const message=choice.message;if(message.role!==void 0&&message.role!=="assistant"||message.function_call!=null||message.refusal!=null||message.content!==null&&typeof message.content!="string"||message.reasoning_content!=null&&typeof message.reasoning_content!="string")throw envelope();const output=typeof message.content=="string"?[{kind:"text",text:message.content}]:[];if(message.tool_calls!==void 0&&message.tool_calls!==null){if(!Array.isArray(message.tool_calls)||message.tool_calls.length>toolLimit)throw envelope();for(const call of message.tool_calls){if(!record(call)||typeof call.id!="string"||call.type!=="function"||!record(call.function)||typeof call.function.name!="string"||typeof call.function.arguments!="string")throw envelope();output.push({kind:"tool-call",id:call.id,name:call.function.name,arguments:{encoding:"json-text",text:call.function.arguments}})}}if(message.content===""&&output.some(item=>item.kind==="tool-call")&&output.shift(),message.content===null&&!output.length&&choice.finish_reason==="stop")throw envelope();const reply={formatVersion:1,id:value.id,model:value.model,output,termination:{kind:terminationKind(choice.finish_reason),reason:choice.finish_reason},usage:acceptUsage(value.usage),native:{format:"chat-completions",version:1,...typeof message.reasoning_content=="string"?{reasoningContent:message.reasoning_content}:{}}};try{return freezeJson(requireModelReply(reply))}catch{throw envelope()}}__name(acceptWireReply,"acceptWireReply");function encodeMessages(request){return[...request.messages.map(message=>({role:message.role,content:message.content})),...(request.exchanges??[]).flatMap(encodeExchange)]}__name(encodeMessages,"encodeMessages");function encodeExchange({reply,feedback}){const calls=reply.output.filter(item=>item.kind==="tool-call"),byId=new Map(feedback.map(item=>[item.toolCallId,item]));return[{role:"assistant",content:reply.output.find(item=>item.kind==="text")?.text??null,...reply.native.reasoningContent===void 0?{}:{reasoning_content:reply.native.reasoningContent},...calls.length?{tool_calls:calls.map(call=>({id:call.id,type:"function",function:{name:call.name,arguments:call.arguments.text}}))}:{}},...calls.map(call=>{const result=byId.get(call.id);return{role:"tool",tool_call_id:call.id,content:JSON.stringify({status:result.status,content:result.content})}})]}__name(encodeExchange,"encodeExchange");function envelope(){return new ProviderError("ENVELOPE","Model response is not a supported Chat Completions reply")}__name(envelope,"envelope");export{acceptWireReply,encodeMessages};

package/package.json

@@ -1,6 +1,40 @@
 {
   "name": "@weiliang0121/capix-model-provider",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
+  "version": "0.2.0",
+  "private": false,
+  "type": "module",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    }
+  },
+  "description": "Chat Completions model protocol adapter with streaming and native tool calls for CAPiX.",
+  "license": "MIT",
+  "author": "wei.liang (https://github.com/weiliang0121)",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/weiliang0121/capix.git",
+    "directory": "packages/model-provider"
+  },
+  "engines": {
+    "node": ">=22.23.2"
+  },
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "files": [
+    "dist/**/*.js",
+    "dist/**/*.d.ts",
+    "LICENSE",
+    "README.md"
+  ],
+  "publishConfig": {
+    "access": "public",
+    "registry": "https://registry.npmjs.org"
+  },
+  "scripts": {
+    "build": "tsc -p tsconfig.json",
+    "typecheck": "tsc -p tsconfig.json --noEmit",
+    "test": "node ../../scripts/test/run.mjs --package ."
+  }
 }