Merge pull request #30 from CJackHwang/dev

feat: Add comprehensive historical and current Claude model IDs for API compatibility and dynamic Docker port configuration.
2026-05-05 00:45:29 +08:00 · 2026-02-17 14:03:03 +08:00
parent 0f389471ac 76ae2fed51
commit a9828e33ad
19 changed files with 347 additions and 105 deletions
--- a/API.en.md
+++ b/API.en.md
@@ -246,13 +246,15 @@ No auth required.
 {
  "object": "list",
  "data": [
-    {"id": "claude-sonnet-4-20250514", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
-    {"id": "claude-sonnet-4-20250514-fast", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
-    {"id": "claude-sonnet-4-20250514-slow", "object": "model", "created": 1715635200, "owned_by": "anthropic"}
+    {"id": "claude-sonnet-4-5", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
+    {"id": "claude-haiku-4-5", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
+    {"id": "claude-opus-4-6", "object": "model", "created": 1715635200, "owned_by": "anthropic"}
  ]
 }
 ```

+> Note: the example is partial; the real response includes historical Claude 1.x/2.x/3.x/4.x IDs and common aliases.
+
 ### `POST /anthropic/v1/messages`

 **Headers**:
@@ -267,7 +269,7 @@ anthropic-version: 2023-06-01

 | Field | Type | Required | Notes |
 | --- | --- | --- | --- |
-| `model` | string | ✅ | `claude-sonnet-4-20250514` / `-fast` / `-slow` |
+| `model` | string | ✅ | For example `claude-sonnet-4-5` / `claude-opus-4-6` / `claude-haiku-4-5` (compatible with `claude-3-5-haiku-latest`), plus historical Claude model IDs |
 | `messages` | array | ✅ | Claude-style messages |
 | `max_tokens` | number | ❌ | Not strictly enforced by upstream bridge |
 | `stream` | boolean | ❌ | Default `false` |
@@ -281,7 +283,7 @@ anthropic-version: 2023-06-01
  "id": "msg_1738400000000000000",
  "type": "message",
  "role": "assistant",
-  "model": "claude-sonnet-4-20250514",
+  "model": "claude-sonnet-4-5",
  "content": [
    {"type": "text", "text": "response"}
  ],
@@ -325,7 +327,7 @@ data: {"type":"message_stop"}

 **Notes**:

- Thinking-enabled models (`-slow`) stream `thinking_delta`
+- Models whose names contain `opus` / `reasoner` / `slow` stream `thinking_delta`
 - `signature_delta` is not emitted (DeepSeek does not provide verifiable thinking signatures)
 - In `tools` mode, the stream avoids leaking raw tool JSON and does not force `input_json_delta`

@@ -335,7 +337,7 @@ data: {"type":"message_stop"}

 ```json
 {
-  "model": "claude-sonnet-4-20250514",
+  "model": "claude-sonnet-4-5",
  "messages": [
    {"role": "user", "content": "Hello"}
  ]
@@ -754,7 +756,7 @@ curl http://localhost:5001/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
-    "model": "claude-sonnet-4-20250514",
+    "model": "claude-sonnet-4-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'
@@ -768,7 +770,7 @@ curl http://localhost:5001/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
-    "model": "claude-sonnet-4-20250514-slow",
+    "model": "claude-opus-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Explain relativity"}],
    "stream": true
--- a/API.md
+++ b/API.md
@@ -246,13 +246,15 @@ data: [DONE]
 {
  "object": "list",
  "data": [
-    {"id": "claude-sonnet-4-20250514", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
-    {"id": "claude-sonnet-4-20250514-fast", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
-    {"id": "claude-sonnet-4-20250514-slow", "object": "model", "created": 1715635200, "owned_by": "anthropic"}
+    {"id": "claude-sonnet-4-5", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
+    {"id": "claude-haiku-4-5", "object": "model", "created": 1715635200, "owned_by": "anthropic"},
+    {"id": "claude-opus-4-6", "object": "model", "created": 1715635200, "owned_by": "anthropic"}
  ]
 }
 ```

+> 说明：示例仅展示部分模型；实际返回包含 Claude 1.x/2.x/3.x/4.x 历史模型 ID 与常见别名。
+
 ### `POST /anthropic/v1/messages`

 **请求头**：
@@ -267,7 +269,7 @@ anthropic-version: 2023-06-01

 | 字段 | 类型 | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `model` | string | ✅ | `claude-sonnet-4-20250514` / `-fast` / `-slow` |
+| `model` | string | ✅ | 例如 `claude-sonnet-4-5` / `claude-opus-4-6` / `claude-haiku-4-5`（兼容 `claude-3-5-haiku-latest`），并支持历史 Claude 模型 ID |
 | `messages` | array | ✅ | Claude 风格消息数组 |
 | `max_tokens` | number | ❌ | 当前实现不会硬性截断上游输出 |
 | `stream` | boolean | ❌ | 默认 `false` |
@@ -281,7 +283,7 @@ anthropic-version: 2023-06-01
  "id": "msg_1738400000000000000",
  "type": "message",
  "role": "assistant",
-  "model": "claude-sonnet-4-20250514",
+  "model": "claude-sonnet-4-5",
  "content": [
    {"type": "text", "text": "回复内容"}
  ],
@@ -325,7 +327,7 @@ data: {"type":"message_stop"}

 **说明**：

- 思维模型（`-slow`）会输出 `thinking_delta`
+- 名称中包含 `opus` / `reasoner` / `slow` 的模型会输出 `thinking_delta`
 - 不会输出 `signature_delta`（上游 DeepSeek 未提供可验证签名）
 - `tools` 场景优先避免泄露原始工具 JSON，不强制发送 `input_json_delta`

@@ -335,7 +337,7 @@ data: {"type":"message_stop"}

 ```json
 {
-  "model": "claude-sonnet-4-20250514",
+  "model": "claude-sonnet-4-5",
  "messages": [
    {"role": "user", "content": "你好"}
  ]
@@ -754,7 +756,7 @@ curl http://localhost:5001/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
-    "model": "claude-sonnet-4-20250514",
+    "model": "claude-sonnet-4-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "你好"}]
  }'
@@ -768,7 +770,7 @@ curl http://localhost:5001/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
-    "model": "claude-sonnet-4-20250514-slow",
+    "model": "claude-opus-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "解释相对论"}],
    "stream": true
--- a/DEPLOY.en.md
+++ b/DEPLOY.en.md
@@ -145,13 +145,20 @@ Docker Compose includes a built-in health check:

 ```yaml
 healthcheck:
-  test: ["CMD", "wget", "-qO-", "http://localhost:5001/healthz"]
+  test: ["CMD", "wget", "-qO-", "http://localhost:${PORT:-5001}/healthz"]
  interval: 30s
  timeout: 10s
  retries: 3
  start_period: 10s
 ```

+### 2.6 Docker Troubleshooting
+
+If container logs look normal but the admin panel is unreachable, check these first:
+
+1. **Port alignment**: when `PORT` is not `5001`, use the same port in your URL (for example `http://localhost:8080/admin`).
+2. **WebUI assets in dev compose**: `docker-compose.dev.yml` runs `go run` in a dev image and does not auto-install Node.js inside the container; if `static/admin` is missing in your repo, `/admin` will return 404. Build once on host: `./scripts/build-webui.sh`.
+
 ---

 ## 3. Vercel Deployment
@@ -211,14 +218,15 @@ Vercel Go Runtime applies platform-level response buffering, so this project use
 1. `api/chat-stream.js` receives `/v1/chat/completions` request
 2. Node calls Go internal prepare endpoint (`?__stream_prepare=1`) for session ID, PoW, token
 3. Go prepare creates a stream lease, locking the account
-4. Node connects directly to DeepSeek upstream, relays SSE in real-time to client
+4. Node connects directly to DeepSeek upstream, relays SSE in real-time to client (including OpenAI chunk framing and tools anti-leak sieve)
 5. After stream ends, Node calls Go release endpoint (`?__stream_release=1`) to free the account

 > This adaptation is **Vercel-only**; local and Docker remain pure Go.

-#### Non-Stream and Tool Call Fallback
+#### Non-Stream Fallback and Tool Call Handling

- `api/chat-stream.js` automatically falls back to Go entry (`?__go=1`) for non-stream requests or requests with `tools`
+- `api/chat-stream.js` falls back to Go entry (`?__go=1`) for non-stream requests only
+- Streaming requests (including requests with `tools`) stay on the Node path and use Go-aligned tool-call anti-leak handling
 - WebUI non-stream test calls `?__go=1` directly to avoid Node hop timeout on long requests

 #### Function Duration
--- a/DEPLOY.md
+++ b/DEPLOY.md
@@ -145,13 +145,20 @@ Docker Compose 已配置内置健康检查：

 ```yaml
 healthcheck:
-  test: ["CMD", "wget", "-qO-", "http://localhost:5001/healthz"]
+  test: ["CMD", "wget", "-qO-", "http://localhost:${PORT:-5001}/healthz"]
  interval: 30s
  timeout: 10s
  retries: 3
  start_period: 10s
 ```

+### 2.6 Docker 常见排查
+
+如果容器日志正常但面板打不开，优先检查：
+
+1. **端口是否一致**：`PORT` 改成非 `5001` 时，访问地址也要改成对应端口（如 `http://localhost:8080/admin`）。
+2. **开发 compose 的 WebUI 静态文件**：`docker-compose.dev.yml` 使用 `go run` 开发镜像，不会在容器内自动安装 Node.js；若仓库里没有 `static/admin`，`/admin` 会返回 404。可先在宿主机构建一次：`./scripts/build-webui.sh`。
+
 ---

 ## 三、Vercel 部署
@@ -211,14 +218,15 @@ api/index.go  api/chat-stream.js
 1. `api/chat-stream.js` 收到 `/v1/chat/completions` 请求
 2. Node 调用 Go 内部 prepare 接口（`?__stream_prepare=1`），获取会话 ID、PoW、token 等
 3. Go prepare 创建 stream lease，锁定账号
-4. Node 直连 DeepSeek 上游，实时流式转发 SSE 给客户端
+4. Node 直连 DeepSeek 上游，实时流式转发 SSE 给客户端（含 OpenAI chunk 封装与 tools 防泄漏筛分）
 5. 流结束后 Node 调用 Go release 接口（`?__stream_release=1`），释放账号

 > 该适配**仅在 Vercel 环境生效**；本地与 Docker 仍走纯 Go 链路。

-#### 非流式与 Tool Call 回退
+#### 非流式回退与 Tool Call 处理

- `api/chat-stream.js` 对非流式请求或带 `tools` 的请求会自动回退到 Go 入口（`?__go=1`）
+- `api/chat-stream.js` 仅对非流式请求回退到 Go 入口（`?__go=1`）
+- 流式请求（包括带 `tools`）走 Node 路径，并执行与 Go 对齐的 tool-call 防泄漏处理
 - WebUI 的"非流式测试"直接请求 `?__go=1`，避免 Node 中转造成长请求超时

 #### 函数时长
--- a/README.MD
+++ b/README.MD
@@ -79,11 +79,12 @@ flowchart LR

 | 模型 | 默认映射 |
 | --- | --- |
-| `claude-sonnet-4-20250514` | `deepseek-chat` |
-| `claude-sonnet-4-20250514-fast` | `deepseek-chat` |
-| `claude-sonnet-4-20250514-slow` | `deepseek-reasoner` |
+| `claude-sonnet-4-5` | `deepseek-chat` |
+| `claude-haiku-4-5`（兼容 `claude-3-5-haiku-latest`） | `deepseek-chat` |
+| `claude-opus-4-6` | `deepseek-reasoner` |

 可通过配置中的 `claude_mapping` 或 `claude_model_mapping` 覆盖映射关系。
+另外，`/anthropic/v1/models` 现已包含 Claude 1.x/2.x/3.x/4.x 历史模型 ID 与常见别名，便于旧客户端直接兼容。

 ## 快速开始

@@ -131,7 +132,7 @@ docker-compose logs -f
 3. 配置环境变量（至少设置 `DS2API_ADMIN_KEY` 和 `DS2API_CONFIG_JSON`）
 4. 部署

-> **流式说明**：`/v1/chat/completions` 在 Vercel 上默认走 `api/chat-stream.js`（Node Runtime）以保证实时 SSE。鉴权、账号选择、会话/PoW 准备仍由 Go 内部 prepare 接口完成，Node 端仅转发流数据。
+> **流式说明**：`/v1/chat/completions` 在 Vercel 上默认走 `api/chat-stream.js`（Node Runtime）以保证实时 SSE。鉴权、账号选择、会话/PoW 准备仍由 Go 内部 prepare 接口完成；流式响应（含 `tools`）在 Node 侧执行与 Go 对齐的输出组装与防泄漏处理。

 详细部署说明请参阅 [部署指南](DEPLOY.md)。

@@ -148,6 +149,22 @@ cp config.example.json config.json
 ./ds2api
 ```

+### 方式五：OpenCode CLI 接入
+
+1. 复制示例配置：
+
+```bash
+cp opencode.json.example opencode.json
+```
+
+2. 编辑 `opencode.json`：
+- 将 `baseURL` 改为你的 DS2API 地址（例如 `https://your-domain.com/v1`）
+- 将 `apiKey` 改为你的 DS2API key（对应 `config.keys`）
+
+3. 在项目目录启动 OpenCode CLI（按你的安装方式运行 `opencode`）。
+
+> 建议优先使用 OpenAI 兼容路径（`/v1/*`），即示例里的 `@ai-sdk/openai-compatible` provider。
+
 ## 配置说明

 ### `config.json` 示例
--- a/README.en.md
+++ b/README.en.md
@@ -79,11 +79,12 @@ flowchart LR

 | Model | Default Mapping |
 | --- | --- |
-| `claude-sonnet-4-20250514` | `deepseek-chat` |
-| `claude-sonnet-4-20250514-fast` | `deepseek-chat` |
-| `claude-sonnet-4-20250514-slow` | `deepseek-reasoner` |
+| `claude-sonnet-4-5` | `deepseek-chat` |
+| `claude-haiku-4-5` (compatible with `claude-3-5-haiku-latest`) | `deepseek-chat` |
+| `claude-opus-4-6` | `deepseek-reasoner` |

 Override mapping via `claude_mapping` or `claude_model_mapping` in config.
+In addition, `/anthropic/v1/models` now includes historical Claude 1.x/2.x/3.x/4.x IDs and common aliases for legacy client compatibility.

 ## Quick Start

@@ -131,7 +132,7 @@ Rebuild after updates: `docker-compose up -d --build`
 3. Set environment variables (minimum: `DS2API_ADMIN_KEY` and `DS2API_CONFIG_JSON`)
 4. Deploy

-> **Streaming note**: `/v1/chat/completions` on Vercel is routed to `api/chat-stream.js` (Node Runtime) for real-time SSE. Auth, account selection, session/PoW preparation are still handled by the Go internal prepare endpoint; Node only relays stream data.
+> **Streaming note**: `/v1/chat/completions` on Vercel is routed to `api/chat-stream.js` (Node Runtime) for real-time SSE. Auth, account selection, and session/PoW preparation are still handled by the Go internal prepare endpoint; streaming output (including `tools`) is assembled on Node with Go-aligned anti-leak handling.

 For detailed deployment instructions, see the [Deployment Guide](DEPLOY.en.md).

@@ -148,6 +149,22 @@ cp config.example.json config.json
 ./ds2api
 ```

+### Option 5: OpenCode CLI
+
+1. Copy the example config:
+
+```bash
+cp opencode.json.example opencode.json
+```
+
+2. Edit `opencode.json`:
+- Set `baseURL` to your DS2API endpoint (for example, `https://your-domain.com/v1`)
+- Set `apiKey` to your DS2API key (from `config.keys`)
+
+3. Start OpenCode CLI in the project directory (run `opencode` using your installed method).
+
+> Recommended: use the OpenAI-compatible path (`/v1/*`) via `@ai-sdk/openai-compatible` as shown in the example.
+
 ## Configuration

 ### `config.json` Example
--- a/api/chat-stream.js
+++ b/api/chat-stream.js
@@ -5,6 +5,7 @@ const {
  createToolSieveState,
  processToolSieveChunk,
  flushToolSieve,
+  parseToolCalls,
  formatOpenAIStreamToolCalls,
 } = require('./helpers/stream-tool-sieve');

@@ -155,20 +156,19 @@ module.exports = async function handler(req, res) {
        return;
      }
      ended = true;
-      if (toolSieveEnabled) {
+      const detected = parseToolCalls(outputText, toolNames);
+      if (detected.length > 0 && !toolCallsEmitted) {
+        toolCallsEmitted = true;
+        sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(detected) });
+      } else if (toolSieveEnabled) {
        const tailEvents = flushToolSieve(toolSieveState, toolNames);
        for (const evt of tailEvents) {
-          if (evt.type === 'tool_calls') {
-            toolCallsEmitted = true;
-            sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(evt.calls) });
-            continue;
-          }
          if (evt.text) {
            sendDeltaFrame({ content: evt.text });
          }
        }
      }
-      if (toolCallsEmitted) {
+      if (detected.length > 0 || toolCallsEmitted) {
        reason = 'tool_calls';
      }
      sendFrame({
@@ -233,8 +233,10 @@ module.exports = async function handler(req, res) {
              continue;
            }
            if (p.type === 'thinking') {
-              thinkingText += p.text;
-              sendDeltaFrame({ reasoning_content: p.text });
+              if (thinkingEnabled) {
+                thinkingText += p.text;
+                sendDeltaFrame({ reasoning_content: p.text });
+              }
            } else {
              outputText += p.text;
              if (!toolSieveEnabled) {
--- a/api/helpers/stream-tool-sieve.js
+++ b/api/helpers/stream-tool-sieve.js
@@ -1,6 +1,7 @@
 'use strict';

 const crypto = require('crypto');
+const TOOL_CALL_PATTERN = /\{\s*["']tool_calls["']\s*:\s*\[(.*?)\]\s*\}/s;

 function extractToolNames(tools) {
  if (!Array.isArray(tools) || tools.length === 0) {
@@ -105,7 +106,7 @@ function flushToolSieve(state, toolNames) {
        events.push({ type: 'text', text: consumed.suffix });
      }
    } else if (state.capture) {
-      events.push({ type: 'text', text: state.capture });
+      // Incomplete captured tool JSON at stream end: suppress raw capture.
    }
    state.capture = '';
    state.capturing = false;
@@ -129,12 +130,9 @@ function splitSafeContentForToolDetection(s) {
  if (suspiciousStart > 0) {
    return [text.slice(0, suspiciousStart), text.slice(suspiciousStart)];
  }
-  const chars = Array.from(text);
-  const maxHold = 128;
-  if (chars.length <= maxHold) {
-    return ['', text];
-  }
-  return [chars.slice(0, chars.length - maxHold).join(''), chars.slice(chars.length - maxHold).join('')];
+  // If suspicious content starts at the beginning, keep holding until we can
+  // either parse a full tool JSON block or reach stream flush.
+  return ['', text];
 }

 function findSuspiciousPrefixStart(s) {
@@ -168,30 +166,23 @@ function consumeToolCapture(captured, toolNames) {
  const lower = captured.toLowerCase();
  const keyIdx = lower.indexOf('tool_calls');
  if (keyIdx < 0) {
-    if (Array.from(captured).length >= 256) {
-      return { ready: true, prefix: captured, calls: [], suffix: '' };
-    }
    return { ready: false, prefix: '', calls: [], suffix: '' };
  }
  const start = captured.slice(0, keyIdx).lastIndexOf('{');
  if (start < 0) {
-    if (Array.from(captured).length >= 512) {
-      return { ready: true, prefix: captured, calls: [], suffix: '' };
-    }
    return { ready: false, prefix: '', calls: [], suffix: '' };
  }
  const obj = extractJSONObjectFrom(captured, start);
  if (!obj.ok) {
-    if (Array.from(captured).length >= 4096) {
-      return { ready: true, prefix: captured, calls: [], suffix: '' };
-    }
    return { ready: false, prefix: '', calls: [], suffix: '' };
  }
  const parsed = parseToolCalls(captured.slice(start, obj.end), toolNames);
  if (parsed.length === 0) {
+    // `tool_calls` key exists but strict JSON parse failed.
+    // Drop the captured object body to avoid leaking raw tool JSON.
    return {
      ready: true,
-      prefix: captured.slice(0, obj.end),
+      prefix: captured.slice(0, start),
      calls: [],
      suffix: captured.slice(obj.end),
    };
@@ -292,24 +283,53 @@ function buildToolCallCandidates(text) {
      candidates.push(toStringSafe(m[1]));
    }
  }
-  const keyIdx = trimmed.toLowerCase().indexOf('tool_calls');
-  if (keyIdx >= 0) {
-    const start = trimmed.slice(0, keyIdx).lastIndexOf('{');
-    if (start >= 0) {
-      const obj = extractJSONObjectFrom(trimmed, start);
-      if (obj.ok) {
-        candidates.push(toStringSafe(trimmed.slice(start, obj.end)));
-      }
-    }
+  for (const candidate of extractToolCallObjects(trimmed)) {
+    candidates.push(toStringSafe(candidate));
  }
  const first = trimmed.indexOf('{');
  const last = trimmed.lastIndexOf('}');
  if (first >= 0 && last > first) {
    candidates.push(toStringSafe(trimmed.slice(first, last + 1)));
  }
+  const m = trimmed.match(TOOL_CALL_PATTERN);
+  if (m && m[1]) {
+    candidates.push(`{"tool_calls":[${m[1]}]}`);
+  }
  return [...new Set(candidates.filter(Boolean))];
 }

+function extractToolCallObjects(text) {
+  const raw = toStringSafe(text);
+  if (!raw) {
+    return [];
+  }
+  const lower = raw.toLowerCase();
+  const out = [];
+  let offset = 0;
+  // eslint-disable-next-line no-constant-condition
+  while (true) {
+    let idx = lower.indexOf('tool_calls', offset);
+    if (idx < 0) {
+      break;
+    }
+    let start = raw.slice(0, idx).lastIndexOf('{');
+    while (start >= 0) {
+      const obj = extractJSONObjectFrom(raw, start);
+      if (obj.ok) {
+        out.push(raw.slice(start, obj.end).trim());
+        offset = obj.end;
+        idx = -1;
+        break;
+      }
+      start = raw.slice(0, start).lastIndexOf('{');
+    }
+    if (idx >= 0) {
+      offset = idx + 'tool_calls'.length;
+    }
+  }
+  return out;
+}
+
 function parseToolCallsPayload(payload) {
  let decoded;
  try {
@@ -452,5 +472,6 @@ module.exports = {
  createToolSieveState,
  processToolSieveChunk,
  flushToolSieve,
+  parseToolCalls,
  formatOpenAIStreamToolCalls,
 };
--- a/docker-compose.dev.yml
+++ b/docker-compose.dev.yml
@@ -16,7 +16,7 @@ services:
    container_name: ds2api-dev
    command: ["go", "run", "./cmd/ds2api"]
    ports:
-      - "${PORT:-5001}:5001"
+      - "${PORT:-5001}:${PORT:-5001}"
    env_file:
      - .env
    environment:
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -4,14 +4,14 @@ services:
    image: ds2api:latest
    container_name: ds2api
    ports:
-      - "${PORT:-5001}:5001"
+      - "${PORT:-5001}:${PORT:-5001}"
    env_file:
      - .env
    environment:
      - HOST=0.0.0.0
    restart: unless-stopped
    healthcheck:
-      test: ["CMD", "wget", "-qO-", "http://localhost:5001/healthz"]
+      test: ["CMD", "wget", "-qO-", "http://localhost:${PORT:-5001}/healthz"]
      interval: 30s
      timeout: 10s
      retries: 3
--- a/internal/adapter/claude/handler_stream_test.go
+++ b/internal/adapter/claude/handler_stream_test.go
@@ -81,7 +81,7 @@ func TestHandleClaudeStreamRealtimeTextIncrementsWithEventHeaders(t *testing.T)
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-20250514", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)

 	body := rec.Body.String()
 	if !strings.Contains(body, "event: message_start") {
@@ -122,7 +122,7 @@ func TestHandleClaudeStreamRealtimeThinkingDelta(t *testing.T) {
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-20250514", []any{map[string]any{"role": "user", "content": "hi"}}, true, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, true, false, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	foundThinkingDelta := false
@@ -148,7 +148,7 @@ func TestHandleClaudeStreamRealtimeToolSafety(t *testing.T) {
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-20250514", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"search"})
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"search"})

 	frames := parseClaudeFrames(t, rec.Body.String())
 	for _, f := range findClaudeFrames(frames, "content_block_delta") {
@@ -191,7 +191,7 @@ func TestHandleClaudeStreamRealtimeUpstreamErrorEvent(t *testing.T) {
 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)

-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-20250514", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	errFrames := findClaudeFrames(frames, "error")
@@ -228,7 +228,7 @@ func TestHandleClaudeStreamRealtimePingEvent(t *testing.T) {

 	rec := httptest.NewRecorder()
 	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)
-	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-20250514", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "hi"}}, false, false, nil)

 	frames := parseClaudeFrames(t, rec.Body.String())
 	if len(findClaudeFrames(frames, "ping")) == 0 {
--- a/internal/adapter/openai/handler_toolcall_test.go
+++ b/internal/adapter/openai/handler_toolcall_test.go
@@ -416,3 +416,124 @@ func TestHandleStreamToolCallMixedWithPlainTextSegments(t *testing.T) {
 		t.Fatalf("expected finish_reason=tool_calls, body=%s", rec.Body.String())
 	}
 }
+
+func TestHandleStreamToolCallKeyAppearsLateStillNoPrefixLeak(t *testing.T) {
+	h := &Handler{}
+	spaces := strings.Repeat(" ", 200)
+	resp := makeSSEHTTPResponse(
+		`data: {"p":"response/content","v":"{`+spaces+`"}`,
+		`data: {"p":"response/content","v":"\"tool_calls\":[{\"name\":\"search\",\"input\":{\"q\":\"go\"}}]}"}`,
+		`data: {"p":"response/content","v":"后置正文C。"}`,
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+
+	h.handleStream(rec, req, resp, "cid8", "deepseek-chat", "prompt", false, false, []string{"search"})
+
+	frames, done := parseSSEDataFrames(t, rec.Body.String())
+	if !done {
+		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
+	}
+	if !streamHasToolCallsDelta(frames) {
+		t.Fatalf("expected tool_calls delta, body=%s", rec.Body.String())
+	}
+	if streamHasRawToolJSONContent(frames) {
+		t.Fatalf("raw tool_calls JSON leaked in content delta: %s", rec.Body.String())
+	}
+	content := strings.Builder{}
+	for _, frame := range frames {
+		choices, _ := frame["choices"].([]any)
+		for _, item := range choices {
+			choice, _ := item.(map[string]any)
+			delta, _ := choice["delta"].(map[string]any)
+			if c, ok := delta["content"].(string); ok {
+				content.WriteString(c)
+			}
+		}
+	}
+	got := content.String()
+	if strings.Contains(got, "{") {
+		t.Fatalf("unexpected suspicious prefix leak in content: %q", got)
+	}
+	if !strings.Contains(got, "后置正文C。") {
+		t.Fatalf("expected stream to continue after tool json convergence, got=%q", got)
+	}
+	if streamFinishReason(frames) != "tool_calls" {
+		t.Fatalf("expected finish_reason=tool_calls, body=%s", rec.Body.String())
+	}
+}
+
+func TestHandleStreamInvalidToolJSONDoesNotLeakRawObject(t *testing.T) {
+	h := &Handler{}
+	resp := makeSSEHTTPResponse(
+		`data: {"p":"response/content","v":"前置正文D。"}`,
+		`data: {"p":"response/content","v":"{'tool_calls':[{'name':'search','input':{'q':'go'}}]}"}`,
+		`data: {"p":"response/content","v":"后置正文E。"}`,
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+
+	h.handleStream(rec, req, resp, "cid9", "deepseek-chat", "prompt", false, false, []string{"search"})
+
+	frames, done := parseSSEDataFrames(t, rec.Body.String())
+	if !done {
+		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
+	}
+	if streamHasToolCallsDelta(frames) {
+		t.Fatalf("did not expect tool_calls delta for invalid json, body=%s", rec.Body.String())
+	}
+	content := strings.Builder{}
+	for _, frame := range frames {
+		choices, _ := frame["choices"].([]any)
+		for _, item := range choices {
+			choice, _ := item.(map[string]any)
+			delta, _ := choice["delta"].(map[string]any)
+			if c, ok := delta["content"].(string); ok {
+				content.WriteString(c)
+			}
+		}
+	}
+	got := strings.ToLower(content.String())
+	if strings.Contains(got, "tool_calls") {
+		t.Fatalf("unexpected raw tool_calls leak in content: %q", content.String())
+	}
+	if !strings.Contains(content.String(), "前置正文D。") || !strings.Contains(content.String(), "后置正文E。") {
+		t.Fatalf("expected pre/post plain text to remain, got=%q", content.String())
+	}
+}
+
+func TestHandleStreamIncompleteCapturedToolJSONDoesNotLeakOnFinalize(t *testing.T) {
+	h := &Handler{}
+	resp := makeSSEHTTPResponse(
+		`data: {"p":"response/content","v":"{\"tool_calls\":[{\"name\":\"search\""}`,
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+
+	h.handleStream(rec, req, resp, "cid10", "deepseek-chat", "prompt", false, false, []string{"search"})
+
+	frames, done := parseSSEDataFrames(t, rec.Body.String())
+	if !done {
+		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
+	}
+	if streamHasToolCallsDelta(frames) {
+		t.Fatalf("did not expect tool_calls delta for incomplete json, body=%s", rec.Body.String())
+	}
+	content := strings.Builder{}
+	for _, frame := range frames {
+		choices, _ := frame["choices"].([]any)
+		for _, item := range choices {
+			choice, _ := item.(map[string]any)
+			delta, _ := choice["delta"].(map[string]any)
+			if c, ok := delta["content"].(string); ok {
+				content.WriteString(c)
+			}
+		}
+	}
+	if strings.Contains(strings.ToLower(content.String()), "tool_calls") || strings.Contains(content.String(), "{") {
+		t.Fatalf("unexpected incomplete tool json leak in content: %q", content.String())
+	}
+}
--- a/internal/adapter/openai/tool_sieve.go
+++ b/internal/adapter/openai/tool_sieve.go
@@ -96,10 +96,7 @@ func flushToolSieve(state *toolStreamSieveState, toolNames []string) []toolStrea
 				events = append(events, toolStreamEvent{Content: consumedSuffix})
 			}
 		} else {
-			raw := state.capture.String()
-			if raw != "" {
-				events = append(events, toolStreamEvent{Content: raw})
-			}
+			// Incomplete captured tool JSON at stream end: suppress raw capture.
 		}
 		state.capture.Reset()
 		state.capturing = false
@@ -122,12 +119,9 @@ func splitSafeContentForToolDetection(s string) (safe, hold string) {
 	if suspiciousStart > 0 {
 		return s[:suspiciousStart], s[suspiciousStart:]
 	}
-	runes := []rune(s)
-	const maxHold = 128
-	if len(runes) <= maxHold {
-		return "", s
-	}
-	return string(runes[:len(runes)-maxHold]), string(runes[len(runes)-maxHold:])
+	// If suspicious content starts at position 0, keep holding until we can
+	// parse a complete tool JSON block or reach stream flush.
+	return "", s
 }

 func findSuspiciousPrefixStart(s string) int {
@@ -167,28 +161,21 @@ func consumeToolCapture(captured string, toolNames []string) (prefix string, cal
 	lower := strings.ToLower(captured)
 	keyIdx := strings.Index(lower, "tool_calls")
 	if keyIdx < 0 {
-		if len([]rune(captured)) >= 256 {
-			return captured, nil, "", true
-		}
 		return "", nil, "", false
 	}
 	start := strings.LastIndex(captured[:keyIdx], "{")
 	if start < 0 {
-		if len([]rune(captured)) >= 512 {
-			return captured, nil, "", true
-		}
 		return "", nil, "", false
 	}
 	obj, end, ok := extractJSONObjectFrom(captured, start)
 	if !ok {
-		if len([]rune(captured)) >= 4096 {
-			return captured, nil, "", true
-		}
 		return "", nil, "", false
 	}
 	parsed := util.ParseToolCalls(obj, toolNames)
 	if len(parsed) == 0 {
-		return captured[:end], nil, captured[end:], true
+		// `tool_calls` key exists but strict JSON parse failed.
+		// Drop the captured object body to avoid leaking raw tool JSON.
+		return captured[:start], nil, captured[end:], true
 	}
 	return captured[:start], parsed, captured[end:], true
 }
--- a/internal/config/models.go
+++ b/internal/config/models.go
@@ -16,9 +16,44 @@ var DeepSeekModels = []ModelInfo{
 }

 var ClaudeModels = []ModelInfo{
+	// Current aliases
+	{ID: "claude-opus-4-6", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-sonnet-4-5", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-haiku-4-5", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+
+	// Current snapshots
+	{ID: "claude-opus-4-5-20251101", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-opus-4-1", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-opus-4-1-20250805", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-opus-4-0", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-opus-4-20250514", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-sonnet-4-5-20250929", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-sonnet-4-0", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
 	{ID: "claude-sonnet-4-20250514", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
-	{ID: "claude-sonnet-4-20250514-fast", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
-	{ID: "claude-sonnet-4-20250514-slow", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-haiku-4-5-20251001", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+
+	// Claude 3.x (legacy/deprecated snapshots and aliases)
+	{ID: "claude-3-7-sonnet-latest", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-7-sonnet-20250219", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-5-sonnet-latest", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-5-sonnet-20240620", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-5-sonnet-20241022", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-opus-20240229", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-sonnet-20240229", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-5-haiku-latest", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-5-haiku-20241022", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-3-haiku-20240307", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+
+	// Claude 2.x and 1.x (retired but accepted for compatibility)
+	{ID: "claude-2.1", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-2.0", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-1.3", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-1.2", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-1.1", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-1.0", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-instant-1.2", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-instant-1.1", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
+	{ID: "claude-instant-1.0", Object: "model", Created: 1715635200, OwnedBy: "anthropic"},
 }

 func GetModelConfig(model string) (thinking bool, search bool, ok bool) {
--- a/internal/testsuite/edge_cases.go
+++ b/internal/testsuite/edge_cases.go
@@ -275,7 +275,7 @@ func (r *Runner) caseSSEJSONIntegrity(ctx context.Context, cc *caseContext) erro
 			"anthropic-version": "2023-06-01",
 		},
 		Body: map[string]any{
-			"model": "claude-sonnet-4-20250514",
+			"model": "claude-sonnet-4-5",
 			"messages": []map[string]any{
 				{"role": "user", "content": "stream json integrity"},
 			},
--- a/internal/testsuite/runner.go
+++ b/internal/testsuite/runner.go
@@ -1056,7 +1056,7 @@ func (r *Runner) caseAnthropicNonstream(ctx context.Context, cc *caseContext) er
 			"content-type":      "application/json",
 		},
 		Body: map[string]any{
-			"model": "claude-sonnet-4-20250514",
+			"model": "claude-sonnet-4-5",
 			"messages": []map[string]any{
 				{"role": "user", "content": "hello"},
 			},
@@ -1084,7 +1084,7 @@ func (r *Runner) caseAnthropicStream(ctx context.Context, cc *caseContext) error
 			"content-type":      "application/json",
 		},
 		Body: map[string]any{
-			"model": "claude-sonnet-4-20250514",
+			"model": "claude-sonnet-4-5",
 			"messages": []map[string]any{
 				{"role": "user", "content": "stream hello"},
 			},
@@ -1113,7 +1113,7 @@ func (r *Runner) caseAnthropicCountTokens(ctx context.Context, cc *caseContext)
 			"content-type":      "application/json",
 		},
 		Body: map[string]any{
-			"model": "claude-sonnet-4-20250514",
+			"model": "claude-sonnet-4-5",
 			"messages": []map[string]any{
 				{"role": "user", "content": "count me"},
 			},
--- a/internal/util/messages.go
+++ b/internal/util/messages.go
@@ -9,7 +9,7 @@ import (

 var markdownImagePattern = regexp.MustCompile(`!\[(.*?)\]\((.*?)\)`)

-const ClaudeDefaultModel = "claude-sonnet-4-20250514"
+const ClaudeDefaultModel = "claude-sonnet-4-5"

 type Message struct {
 	Role    string `json:"role"`
--- a/internal/util/messages_test.go
+++ b/internal/util/messages_test.go
@@ -36,7 +36,7 @@ func TestMessagesPrepareRoles(t *testing.T) {
 func TestConvertClaudeToDeepSeek(t *testing.T) {
 	store := config.LoadStore()
 	req := map[string]any{
-		"model":    "claude-sonnet-4-20250514-slow",
+		"model":    "claude-opus-4-6",
 		"messages": []any{map[string]any{"role": "user", "content": "Hi"}},
 		"system":   "You are helpful",
 		"stream":   true,
--- a/opencode.json.example
+++ b/opencode.json.example
@@ -0,0 +1,22 @@
+{
+  "$schema": "https://opencode.ai/config.json",
+  "provider": {
+    "ds2api": {
+      "npm": "@ai-sdk/openai-compatible",
+      "name": "DS2API",
+      "options": {
+        "baseURL": "http://localhost:5001/v1",
+        "apiKey": "your-api-key"
+      },
+      "models": {
+        "deepseek-chat": {
+          "name": "DeepSeek Chat (DS2API)"
+        },
+        "deepseek-reasoner": {
+          "name": "DeepSeek Reasoner (DS2API)"
+        }
+      }
+    }
+  },
+  "model": "ds2api/deepseek-chat"
+}