Update CLI chat session bridge (#697)

* feat: add CLI chat sessions with Python agent bridge Introduce a new CLI chat mode that connects Web UI directly to Hermes Agent's AIAgent via a Python bridge subprocess and Socket.IO, bypassing the API Server /v1/responses path. Supports streaming, slash commands (/new, /undo, /retry, /branch, /compress, /save, /title), interrupt, and steer. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * feat: update CLI chat session bridge * fix: extend agent bridge startup timeouts * docs: update bridge chat session design * feat: align bridge compression and provider registry * chore: bump version to 0.5.20 --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-14 09:03:57 +08:00
parent e0fcc0040b
commit eae7195ba8
31 changed files with 3906 additions and 1040 deletions
@@ -10,6 +10,8 @@ package-lock.json
 node_modules
 dist
 dist-ssr
+__pycache__/
+*.py[cod]
 server/dist
 packages/server/dist
 *.local
@@ -37,4 +39,4 @@ hermes-dependencies.md
 .superpowers/
 CLAUDE.md
 # Client source map artifacts
-packages/client/src/**/*.js
+packages/client/src/**/*.js
@@ -33,6 +33,7 @@ RUN npm run build && npm prune --omit=dev
 ENV NODE_ENV=production
 ENV HOME=/home/agent
 ENV HERMES_HOME=/home/agent/.hermes
+ENV PATH=/opt/hermes/.venv/bin:$PATH

 EXPOSE 6060

@@ -14,6 +14,7 @@ services:
      - PORT=${PORT:-6060}
      - HERMES_HOME=/home/agent/.hermes
      - HERMES_BIN=/opt/hermes/.venv/bin/hermes
+      - PATH=/opt/hermes/.venv/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
      - AUTH_DISABLED=${AUTH_DISABLED:-false}
      - HERMES_ALLOW_ROOT_GATEWAY=1
    restart: unless-stopped
@@ -0,0 +1,459 @@
+# CLI/Bridge Chat Sessions 实现文档
+
+> 分支：`feat/cli-chat-sessions`
+
+## 概述
+
+当前实现把原来的聊天通道统一到 Socket.IO namespace `/chat-run`。前端仍使用同一套 `ChatPanel + MessageList + ChatInput`，通过会话的 `source` 字段区分运行方式：
+
+| source | 运行路径 | 说明 |
+|--------|----------|------|
+| `api_server` | Web UI Server → Hermes Gateway `/v1/responses` | 默认聊天路径 |
+| `cli` | Web UI Server → Python agent bridge → `AIAgent` | Bridge(beta)，在 Web UI 服务端子进程里直接运行 Hermes Agent |
+
+Bridge 会话不是一个独立 UI 面板，而是普通会话的一种来源。用户通过“新建聊天”下拉菜单选择 `API` 或 `Bridge (beta)`。
+
+Bridge 模式支持：
+
+- 流式文本输出
+- reasoning/thinking 增量
+- tool started/completed 事件
+- 工具审批请求与响应
+- abort 中断
+- per-session 队列
+- profile 隔离
+- 从 DB resume 会话
+- 与 API Server 路径共用上下文压缩逻辑
+
+当前不再支持旧文档里的独立 `/cli-chat-run` namespace、`CliChatPanel.vue`、`cli-chat.ts` 和 CLI 命令控制层。前端不会再发送 `command` 或 `steer` socket 事件，也不会把 `/new`、`/reset`、`/undo`、`/retry`、`/branch`、`/compress` 等输入当作特殊命令处理。
+
+---
+
+## 整体架构
+
+```text
+ChatPanel.vue
+  ├─ MessageList.vue
+  └─ ChatInput.vue
+        │
+        │ Socket.IO /chat-run
+        ▼
+ChatRunSocket (Node.js)
+  ├─ source=api_server → Hermes Gateway /v1/responses
+  └─ source=cli        → AgentBridgeClient
+                              │ TCP/Unix socket, newline JSON
+                              ▼
+                         hermes_bridge.py
+                              │ in-process import
+                              ▼
+                         AIAgent (hermes-agent)
+```
+
+### 分流规则
+
+`ChatRunSocket.resolveRunSource()` 决定本轮运行走哪个后端：
+
+1. `run` payload 中 `source === 'cli'` 时走 bridge。
+2. `source === 'api_server'` 时走 gateway。
+3. 未显式传 `source` 时，如果 DB 中已有 session 的 `source` 是 `cli`，继续走 bridge。
+4. 其他情况默认走 `api_server`。
+
+---
+
+## 主要文件
+
+### 前端
+
+| 文件 | 说明 |
+|------|------|
+| `packages/client/src/components/hermes/chat/ChatPanel.vue` | 统一聊天面板；新建菜单包含 `API` 和 `Bridge (beta)`；渲染审批条 |
+| `packages/client/src/components/hermes/chat/MessageList.vue` | 统一消息列表；展示文本、reasoning、tool 消息等 |
+| `packages/client/src/components/hermes/chat/ChatInput.vue` | 统一输入框；发送、停止、附件上传入口 |
+| `packages/client/src/api/hermes/chat.ts` | `/chat-run` Socket.IO 客户端；注册 session 事件处理器；发送 run/abort/approval |
+| `packages/client/src/stores/hermes/chat.ts` | 会话状态、发送流程、resume、队列、审批、消息映射 |
+
+### 后端
+
+| 文件 | 说明 |
+|------|------|
+| `packages/server/src/services/hermes/chat-run-socket.ts` | `/chat-run` Socket.IO 服务；同时处理 API Server 和 Bridge 运行 |
+| `packages/server/src/services/hermes/agent-bridge/client.ts` | Node 端 bridge 客户端；通过 socket 请求 Python bridge |
+| `packages/server/src/services/hermes/agent-bridge/manager.ts` | Python bridge 子进程生命周期管理 |
+| `packages/server/src/services/hermes/agent-bridge/hermes_bridge.py` | Python bridge 服务；创建并复用 `AIAgent` 实例 |
+| `packages/server/src/services/hermes/agent-bridge/index.ts` | bridge 模块导出 |
+| `packages/server/src/index.ts` | 启动 `AgentBridgeManager` 和 `ChatRunSocket` |
+| `packages/server/src/services/shutdown.ts` | 关闭时停止 chat socket 和 bridge 子进程 |
+| `packages/server/src/controllers/hermes/sessions.ts` | 会话列表和详情读取，包含 `source` 信息 |
+| `packages/server/src/controllers/hermes/profiles.ts` | profile 切换/管理时清理 bridge 内存会话 |
+
+### 已移除的旧文件
+
+| 文件 | 状态 |
+|------|------|
+| `packages/client/src/api/hermes/cli-chat.ts` | 已删除 |
+| `packages/client/src/components/hermes/chat/CliChatPanel.vue` | 已删除 |
+| `packages/server/src/services/hermes/cli-chat-run-socket.ts` | 已删除 |
+
+---
+
+## 前端流程
+
+### 新建会话
+
+`ChatPanel.vue` 中的新建按钮使用下拉菜单：
+
+- `API`：调用 `chatStore.newChat()`，创建默认 `api_server` 会话。
+- `Bridge (beta)`：调用 `chatStore.newCliSession()`，创建 `source: 'cli'` 会话。
+
+Bridge 会话 ID 使用类似 `YYYYMMDD_HHMMSS_xxxxxx` 的格式，便于与 Hermes CLI 风格的 session ID 对齐。
+
+### 发送消息
+
+1. `ChatInput.vue` 触发 store 的发送逻辑。
+2. `chat.ts` 根据 active session 组装输入内容，附件会被转为 `ContentBlock[]`。
+3. 调用 `startRunViaSocket()`。
+4. 前端向 `/chat-run` emit：
+
+```ts
+socket.emit('run', {
+  session_id,
+  input,
+  instructions,
+  model,
+  queue_id,
+  source, // api_server 或 cli
+})
+```
+
+5. 前端注册本 session 的事件 handler，通过 `session_id` 隔离多会话并发事件。
+
+### Resume
+
+切换会话、页面恢复可见、或刷新后，前端通过：
+
+```ts
+socket.emit('resume', { session_id })
+```
+
+服务端返回：
+
+```ts
+{
+  session_id,
+  messages,
+  isWorking,
+  isAborting,
+  events,
+  inputTokens,
+  outputTokens,
+  queueLength,
+}
+```
+
+如果服务端发现该 session 仍在运行，前端会重新注册 handler，并允许继续 abort。
+
+### 审批
+
+Bridge 工具需要人工确认时，服务端会发 `approval.requested`，前端 store 记录为 `activePendingApproval`，`ChatPanel.vue` 在输入框上方显示审批条。
+
+前端响应审批：
+
+```ts
+socket.emit('approval.respond', {
+  session_id,
+  approval_id,
+  choice, // once | session | always | deny
+})
+```
+
+---
+
+## `/chat-run` Socket.IO 协议
+
+### 客户端 → 服务端
+
+| 事件 | 数据 | 说明 |
+|------|------|------|
+| `run` | `{ session_id, input, model?, instructions?, queue_id?, source? }` | 启动一轮运行；`source` 决定 API Server 或 Bridge |
+| `resume` | `{ session_id }` | 加入 session room 并恢复状态 |
+| `abort` | `{ session_id }` | 中断当前运行 |
+| `cancel_queued_run` | `{ session_id, queue_id }` | 取消等待队列中的一条 run |
+| `approval.respond` | `{ session_id, approval_id, choice }` | 响应 Bridge 工具审批 |
+
+当前没有 `command`、`steer` 或 slash-command 相关 Socket.IO 事件。
+
+### 服务端 → 客户端
+
+| 事件 | 说明 |
+|------|------|
+| `resumed` | 返回 DB 消息、运行状态、队列长度和最近事件 |
+| `run.started` | 运行开始 |
+| `run.queued` | 当前 session 已有运行，新请求进入队列 |
+| `message.delta` | 文本增量 |
+| `reasoning.delta` | reasoning 增量 |
+| `thinking.delta` | thinking 增量 |
+| `reasoning.available` | reasoning 内容可用 |
+| `tool.started` | 工具调用开始 |
+| `tool.completed` | 工具调用结束 |
+| `approval.requested` | Bridge 工具请求人工审批 |
+| `approval.resolved` | 审批完成或超时 |
+| `compression.started` | 上下文压缩开始 |
+| `compression.completed` | 上下文压缩结束 |
+| `usage.updated` | token 用量更新 |
+| `abort.started` | 中断开始 |
+| `abort.completed` | 中断结束 |
+| `run.completed` | 运行完成 |
+| `run.failed` | 运行失败 |
+
+### 认证
+
+`/chat-run` 使用 Socket.IO auth token：
+
+```ts
+io(`${baseUrl}/chat-run`, {
+  auth: { token },
+  query: { profile },
+})
+```
+
+如果未设置 `AUTH_DISABLED=1`，服务端会与 Web UI token 比对。
+
+---
+
+## ChatRunSocket 后端行为
+
+### API Server 路径
+
+`source=api_server` 时：
+
+1. 写入用户消息到 Web UI 本地 session DB。
+2. 通过 `buildCompressedHistory()` 构建上下文。
+3. 请求当前 profile 的 Hermes Gateway：
+
+```text
+POST <upstream>/v1/responses
+```
+
+4. 读取 SSE frame，映射为统一的 `/chat-run` 事件。
+5. 完成后写入 assistant/tool 消息，更新 usage。
+
+### Bridge 路径
+
+`source=cli` 时：
+
+1. 写入用户消息到 Web UI 本地 session DB。
+2. 复用同一套 `buildCompressedHistory()` 构建压缩上下文。
+3. 调用：
+
+```ts
+this.bridge.chat(session_id, input, history, instructions, profile)
+```
+
+4. 轮询 `AgentBridgeClient.streamOutput(run_id)`。
+5. 将 Python bridge 的 delta 和 events 映射成统一事件。
+6. 将 assistant 文本、reasoning、tool 调用结果 flush 回 DB。
+
+### 队列
+
+同一个 `session_id` 同时只能有一个 active run。新的 `run` 到达时：
+
+- 如果当前 session 正在运行，则放入 `state.queue`。
+- 发送 `run.queued` 更新队列长度。
+- 当前 run 结束或 abort 完成后，自动执行下一条 queued run。
+
+---
+
+## Python Agent Bridge
+
+### 通信协议
+
+Node 和 Python bridge 之间使用本地 socket 的单行 JSON 协议：
+
+```json
+{ "action": "chat", "session_id": "xxx", "message": "hello" }
+```
+
+响应也是单行 JSON：
+
+```json
+{ "ok": true, "run_id": "xxx", "session_id": "xxx", "status": "running" }
+```
+
+### Endpoint
+
+默认 endpoint 按平台选择：
+
+| 平台 | 默认 endpoint |
+|------|---------------|
+| Windows | `tcp://127.0.0.1:18765` |
+| macOS/Linux | `ipc:///tmp/hermes-agent-bridge.sock` |
+
+Windows 使用 TCP 是因为部分 Python/Windows 环境没有 Unix domain socket 支持。
+
+### 当前实际使用的 action
+
+| Action | 说明 |
+|--------|------|
+| `chat` | 启动一轮 `AIAgent.run_conversation()` |
+| `get_output` | 通过 `cursor` 和 `event_cursor` 获取增量文本与事件 |
+| `interrupt` | 调用 agent 中断当前运行 |
+| `approval_respond` | 响应工具审批 |
+| `destroy_all` | profile 切换/管理时销毁全部 bridge 内存 session |
+
+bridge 代码里还保留了一些调试/维护 action，例如 `ping`、`get_result`、`get_history`、`destroy`、`list`、`shutdown`、`steer`，但当前 `/chat-run` 前端路径不会暴露这些能力。
+
+旧的 `command` action 已移除，bridge 不再处理 `/new`、`/undo`、`/retry`、`/branch`、`/compress` 等斜杠命令。
+
+### 会话和 profile
+
+`AgentPool` 维护 `session_id -> AgentSession`：
+
+- 每个 session 持有独立 `AIAgent` 实例。
+- session 按 profile 创建，profile 改变时会重建对应 agent。
+- `HERMES_HOME` 会在创建 agent 时临时切到 profile home。
+- `SessionDB` 按 profile 的 `state.db` 路径缓存。
+- 空闲 session 会被 bridge GC，默认 30 分钟无运行后销毁内存态。
+
+### 工具和审批事件
+
+bridge 从 `AIAgent` 回调中收集事件：
+
+- `stream.delta`
+- `reasoning.delta`
+- `thinking.delta`
+- `tool.started`
+- `tool.completed`
+- `tool.progress`
+- `approval.requested`
+- `approval.resolved`
+- `turn.boundary`
+- `status`
+
+`ChatRunSocket` 会把这些事件转换为前端统一事件，并负责 DB 落盘。
+
+审批默认等待 60 秒，超时自动 `deny`。
+
+---
+
+## AgentBridgeClient
+
+`AgentBridgeClient` 是 Node 端本地 socket 客户端。
+
+行为：
+
+- 支持 `ipc://` 和 `tcp://` endpoint。
+- 每次请求新建 socket，发送一行 JSON，读取一行 JSON。
+- 请求通过内部 lock 串行化。
+- 默认请求响应超时为 `120000ms`。
+- `streamOutput()` 每 100ms 轮询一次 `get_output`。
+
+示例：
+
+```ts
+const started = await bridge.chat(sessionId, input, history, instructions, profile)
+
+for await (const chunk of bridge.streamOutput(started.run_id)) {
+  // chunk.delta
+  // chunk.events
+  // chunk.done
+}
+```
+
+注意：目前 socket connect 阶段没有独立 connect timeout，主要依赖系统连接错误和请求响应 timeout。
+
+---
+
+## AgentBridgeManager
+
+`AgentBridgeManager` 负责启动和停止 Python bridge。
+
+启动流程：
+
+1. 定位 `hermes_bridge.py`。
+2. 发现 `hermes-agent` 根目录。
+3. 选择 Python 解释器。
+4. 以子进程启动：
+
+```text
+python hermes_bridge.py --endpoint <endpoint> --agent-root <root> --hermes-home <home>
+```
+
+5. 监听 stdout，等待：
+
+```json
+{ "event": "ready", "endpoint": "..." }
+```
+
+6. 默认 ready 超时为 `120000ms`。
+
+Python 选择优先级：
+
+1. `HERMES_AGENT_BRIDGE_PYTHON`
+2. `agentRoot/venv` 或 `agentRoot/.venv`
+3. installed `hermes` 命令 shebang
+4. `uv run --project <agentRoot> python`
+5. 系统 `python3` / `python`
+
+关闭时先发 `SIGTERM`，1.5 秒后仍未退出则 `SIGKILL`。
+
+---
+
+## 启动与关闭
+
+### 启动
+
+`bootstrap()` 中会先尝试启动 bridge：
+
+```ts
+agentBridgeManager = await startAgentBridgeManager()
+```
+
+bridge 启动失败不会阻止 Web UI 启动，但 Bridge(beta) 会话后续运行会失败。
+
+随后创建统一的 chat socket：
+
+```ts
+chatRunServer = new ChatRunSocket(groupChatServer.getIO(), getGatewayManagerInstance())
+chatRunServer.init()
+```
+
+### 关闭
+
+服务关闭时会清理：
+
+- `/chat-run` Socket.IO 状态
+- Python agent bridge 子进程
+- 其他 WebSocket/Socket.IO 服务
+
+---
+
+## 环境变量
+
+| 变量 | 说明 |
+|------|------|
+| `HERMES_AGENT_BRIDGE_ENDPOINT` | Bridge endpoint；Windows 默认 `tcp://127.0.0.1:18765`，macOS/Linux 默认 `ipc:///tmp/hermes-agent-bridge.sock` |
+| `HERMES_AGENT_BRIDGE_TIMEOUT_MS` | Node 等待 bridge 请求响应的超时，默认 `120000` ms |
+| `HERMES_AGENT_BRIDGE_STARTUP_TIMEOUT_MS` | Node 等待 Python bridge ready 的超时，默认 `120000` ms |
+| `HERMES_AGENT_BRIDGE_PYTHON` | 指定 Python 解释器路径 |
+| `HERMES_AGENT_ROOT` | hermes-agent 安装目录 |
+| `HERMES_AGENT_BRIDGE_UV` | 指定 uv 可执行文件路径 |
+| `HERMES_AGENT_BRIDGE_PLATFORM` | bridge 传给 Hermes Agent 的平台标识，默认 `cli` |
+| `HERMES_BRIDGE_PROVIDER` | 覆盖 bridge 使用的 provider |
+| `HERMES_BRIDGE_MAX_TURNS` | 覆盖 bridge 最大轮数 |
+| `UV` | uv 可执行文件路径 fallback |
+
+Windows 首次启动慢时可以临时放大：
+
+```powershell
+$env:HERMES_AGENT_BRIDGE_STARTUP_TIMEOUT_MS = "300000"
+$env:HERMES_AGENT_BRIDGE_TIMEOUT_MS = "300000"
+```
+
+---
+
+## 当前限制
+
+- Bridge(beta) 仍依赖 Python bridge 成功启动；启动失败时 Web UI 可用，但 bridge 会话不可用。
+- bridge socket connect 阶段还没有单独 connect timeout。
+- 旧 CLI 独立面板和独立 `/cli-chat-run` namespace 已移除。
+- 旧 bridge 斜杠命令和 `command/steer` socket 控制层已移除；现在输入框内容一律按普通用户消息发送。
@@ -1,6 +1,6 @@
 {
  "name": "hermes-web-ui",
-  "version": "0.5.17",
+  "version": "0.5.20",
  "description": "Self-hosted AI chat dashboard for Hermes Agent — multi-model web UI with multi-platform integration",
  "repository": {
    "type": "git",
@@ -17,6 +17,7 @@ export interface StartRunRequest {
  session_id?: string
  model?: string
  queue_id?: string
+  source?: 'api_server' | 'cli'
 }

 export interface StartRunResponse {
@@ -77,6 +78,8 @@ const sessionEventHandlers = new Map<string, {
  onAbortCompleted: (event: RunEvent) => void
  onUsageUpdated: (event: RunEvent) => void
  onRunQueued?: (event: RunEvent) => void
+  onApprovalRequested?: (event: RunEvent) => void
+  onApprovalResolved?: (event: RunEvent) => void
 }>()

 /**
@@ -288,6 +291,26 @@ function globalUsageUpdatedHandler(event: RunEvent): void {
  }
 }

+function globalApprovalRequestedHandler(event: RunEvent): void {
+  const sid = event.session_id
+  if (!sid) return
+
+  const handlers = sessionEventHandlers.get(sid)
+  if (handlers?.onApprovalRequested) {
+    handlers.onApprovalRequested(event)
+  }
+}
+
+function globalApprovalResolvedHandler(event: RunEvent): void {
+  const sid = event.session_id
+  if (!sid) return
+
+  const handlers = sessionEventHandlers.get(sid)
+  if (handlers?.onApprovalResolved) {
+    handlers.onApprovalResolved(event)
+  }
+}
+
 /**
 * Register event handlers for a session
 * @param sessionId - Session ID
@@ -312,6 +335,8 @@ export function registerSessionHandlers(
    onAbortCompleted: (event: RunEvent) => void
    onUsageUpdated: (event: RunEvent) => void
    onRunQueued?: (event: RunEvent) => void
+    onApprovalRequested?: (event: RunEvent) => void
+    onApprovalResolved?: (event: RunEvent) => void
  }
 ): () => void {
  sessionEventHandlers.set(sessionId, handlers)
@@ -330,6 +355,19 @@ export function unregisterSessionHandlers(sessionId: string): void {
  sessionEventHandlers.delete(sessionId)
 }

+export function respondToolApproval(
+  sessionId: string,
+  approvalId: string,
+  choice: 'once' | 'session' | 'always' | 'deny',
+): void {
+  const socket = connectChatRun()
+  socket.emit('approval.respond', {
+    session_id: sessionId,
+    approval_id: approvalId,
+    choice,
+  })
+}
+
 export function getChatRunSocket(): Socket | null {
  return chatRunSocket
 }
@@ -365,7 +403,9 @@ export function connectChatRun(): Socket {
    reconnection: true,
    reconnectionAttempts: Infinity,
    reconnectionDelay: 1000,
-    reconnectionDelayMax: 10000,
+    reconnectionDelayMax: 30000,
+    randomizationFactor: 0.5,
+    timeout: 30000,
  })

  // Register global listeners only once per socket connection
@@ -385,6 +425,8 @@ export function connectChatRun(): Socket {
    chatRunSocket.on('run.failed', globalRunFailedHandler)
    chatRunSocket.on('run.completed', globalRunCompletedHandler)
    chatRunSocket.on('run.queued', globalRunQueuedHandler)
+    chatRunSocket.on('approval.requested', globalApprovalRequestedHandler)
+    chatRunSocket.on('approval.resolved', globalApprovalResolvedHandler)

    // Compression events
    chatRunSocket.on('compression.started', globalCompressionStartedHandler)
@@ -527,6 +569,14 @@ export function startRunViaSocket(
      if (closed) return
      onEvent(evt)
    },
+    onApprovalRequested: (evt: RunEvent) => {
+      if (closed) return
+      onEvent(evt)
+    },
+    onApprovalResolved: (evt: RunEvent) => {
+      if (closed) return
+      onEvent(evt)
+    },
  }

  // Register handlers in the global session map
@@ -66,11 +66,13 @@ export function connectGroupChat(opts?: { userId?: string; userName?: string; de
            name: opts?.userName || localStorage.getItem('gc_user_name') || undefined,
            description: opts?.description || localStorage.getItem('gc_user_description') || undefined,
        },
-        transports: ['websocket'],
+        transports: ['websocket', 'polling'],
        reconnection: true,
        reconnectionAttempts: Infinity,
        reconnectionDelay: 1000,
        reconnectionDelayMax: 30000,
+        randomizationFactor: 0.5,
+        timeout: 30000,
    })

    return socket
@@ -185,4 +187,3 @@ export async function forceCompress(roomId: string): Promise<{ success: boolean;
        method: 'POST',
    })
 }
-
@@ -124,11 +124,16 @@ const groupedSessions = computed<SessionGroup[]>(() => {

  return keys.map((key) => ({
    source: key,
-    label: key ? getSourceLabel(key) : t("chat.other"),
+    label: key ? getChatSourceLabel(key) : t("chat.other"),
    sessions: sortSessionsWithActiveFirst(map.get(key)!),
  }));
 });

+function getChatSourceLabel(source?: string): string {
+  if (source === "cli") return "Bridge (beta)";
+  return getSourceLabel(source);
+}
+
 function toggleGroup(source: string) {
  const isExpanded = !collapsedGroups.value.has(source);
  if (isExpanded) {
@@ -204,10 +209,40 @@ const activeSessionSource = computed(() =>
  currentMode.value === "chat" ? chatStore.activeSession?.source || "" : "",
 );

+const activeApproval = computed(() => chatStore.activePendingApproval);
+
 function handleNewChat() {
  chatStore.newChat();
 }

+function handleNewCliChat() {
+  const session = chatStore.newCliSession()
+  chatStore.switchSession(session.id)
+}
+
+const newChatOptions = computed(() => [
+  {
+    label: "API",
+    key: "api_server",
+  },
+  {
+    label: "Bridge (beta)",
+    key: "cli",
+  },
+]);
+
+function handleNewChatSelect(key: string | number) {
+  if (key === "cli") {
+    handleNewCliChat();
+    return;
+  }
+  handleNewChat();
+}
+
+function handleApproval(choice: "once" | "session" | "always" | "deny") {
+  chatStore.respondApproval(choice);
+}
+
 async function copySessionId(id?: string) {
  const sessionId = id || chatStore.activeSessionId;
  if (sessionId) {
@@ -556,21 +591,27 @@ async function handleWorkspaceConfirm() {
              </svg>
            </template>
          </NButton>
-          <NButton quaternary size="tiny" @click="handleNewChat" circle>
-            <template #icon>
-              <svg
-                width="14"
-                height="14"
-                viewBox="0 0 24 24"
-                fill="none"
-                stroke="currentColor"
-                stroke-width="2"
-              >
-                <line x1="12" y1="5" x2="12" y2="19" />
-                <line x1="5" y1="12" x2="19" y2="12" />
-              </svg>
-            </template>
-          </NButton>
+          <NDropdown
+            trigger="click"
+            :options="newChatOptions"
+            @select="handleNewChatSelect"
+          >
+            <NButton quaternary size="tiny" circle>
+              <template #icon>
+                <svg
+                  width="14"
+                  height="14"
+                  viewBox="0 0 24 24"
+                  fill="none"
+                  stroke="currentColor"
+                  stroke-width="2"
+                >
+                  <line x1="12" y1="5" x2="12" y2="19" />
+                  <line x1="5" y1="12" x2="19" y2="12" />
+                </svg>
+              </template>
+            </NButton>
+          </NDropdown>
        </div>
      </div>
      <div v-if="showSessions" class="session-scope-note">
@@ -723,7 +764,7 @@ async function handleWorkspaceConfirm() {
          </NButton>
          <span class="header-session-title">{{ headerTitle }}</span>
          <span v-if="activeSessionSource" class="source-badge">{{
-            getSourceLabel(activeSessionSource)
+            getChatSourceLabel(activeSessionSource)
          }}</span>
          <span
            v-if="chatStore.activeSession?.workspace"
@@ -766,28 +807,74 @@ async function handleWorkspaceConfirm() {
              </template>
              {{ t("chat.copySessionId") }}
            </NTooltip>
-            <NButton size="small" :circle="isMobile" @click="handleNewChat">
-              <template #icon>
-                <svg
-                  width="14"
-                  height="14"
-                  viewBox="0 0 24 24"
-                  fill="none"
-                  stroke="currentColor"
-                  stroke-width="2"
-                >
-                  <line x1="12" y1="5" x2="12" y2="19" />
-                  <line x1="5" y1="12" x2="19" y2="12" />
-                </svg>
-              </template>
-              <template v-if="!isMobile">{{ t("chat.newChat") }}</template>
-            </NButton>
+            <NDropdown
+              trigger="click"
+              :options="newChatOptions"
+              @select="handleNewChatSelect"
+            >
+              <NButton size="small" :circle="isMobile">
+                <template #icon>
+                  <svg
+                    width="14"
+                    height="14"
+                    viewBox="0 0 24 24"
+                    fill="none"
+                    stroke="currentColor"
+                    stroke-width="2"
+                  >
+                    <line x1="12" y1="5" x2="12" y2="19" />
+                    <line x1="5" y1="12" x2="19" y2="12" />
+                  </svg>
+                </template>
+                <template v-if="!isMobile">{{ t("chat.newChat") }}</template>
+              </NButton>
+            </NDropdown>
          </template>
        </div>
      </header>

      <template v-if="currentMode === 'chat'">
        <MessageList />
+        <div v-if="activeApproval" class="approval-bar">
+          <div class="approval-main">
+            <div class="approval-title">Tool approval required</div>
+            <div class="approval-desc">{{ activeApproval.description }}</div>
+            <code class="approval-command">{{ activeApproval.command }}</code>
+          </div>
+          <div class="approval-actions">
+            <NButton
+              v-if="activeApproval.choices.includes('once')"
+              size="small"
+              type="primary"
+              @click="handleApproval('once')"
+            >
+              Allow once
+            </NButton>
+            <NButton
+              v-if="activeApproval.choices.includes('session')"
+              size="small"
+              @click="handleApproval('session')"
+            >
+              Allow session
+            </NButton>
+            <NButton
+              v-if="activeApproval.choices.includes('always')"
+              size="small"
+              @click="handleApproval('always')"
+            >
+              Always
+            </NButton>
+            <NButton
+              v-if="activeApproval.choices.includes('deny')"
+              size="small"
+              type="error"
+              ghost
+              @click="handleApproval('deny')"
+            >
+              Deny
+            </NButton>
+          </div>
+        </div>
        <ChatInput />
      </template>
      <ConversationMonitorPane
@@ -1259,6 +1346,54 @@ async function handleWorkspaceConfirm() {
  }
 }

+.approval-bar {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  padding: 10px 16px;
+  border-top: 1px solid $border-color;
+  background: $bg-card;
+}
+
+.approval-main {
+  flex: 1;
+  min-width: 0;
+}
+
+.approval-title {
+  font-size: 13px;
+  font-weight: 600;
+  color: $text-primary;
+}
+
+.approval-desc {
+  margin-top: 2px;
+  font-size: 12px;
+  color: $text-secondary;
+}
+
+.approval-command {
+  display: block;
+  margin-top: 6px;
+  max-height: 56px;
+  overflow: auto;
+  white-space: pre-wrap;
+  word-break: break-word;
+  font-size: 12px;
+  color: $text-primary;
+  background: $bg-secondary;
+  border: 1px solid $border-color;
+  border-radius: 6px;
+  padding: 6px 8px;
+}
+
+.approval-actions {
+  display: flex;
+  flex-wrap: wrap;
+  justify-content: flex-end;
+  gap: 6px;
+}
+
@keyframes rainbow-glow {
  0% {
    box-shadow:
@@ -26,10 +26,6 @@ function formatToolDuration(seconds: number): string {
  return `${mins}m ${secs}s`
 }

-const displayMessages = computed(() =>
-  chatStore.messages.filter((m) => m.role !== "tool"),
-);
-
 const currentToolCalls = computed(() => {
  const msgs = chatStore.messages;
  // Find the last user message index
@@ -45,6 +41,22 @@ const currentToolCalls = computed(() => {
  return [...tools].reverse();
 });

+const displayMessages = computed(() =>
+  chatStore.messages.filter((m) => {
+    if (m.role === "tool") return false;
+    if (
+      m.role === "assistant" &&
+      m.isStreaming &&
+      !m.content?.trim() &&
+      !!m.reasoning?.trim() &&
+      currentToolCalls.value.length === 0
+    ) {
+      return false;
+    }
+    return true;
+  }),
+);
+
 const queuedMessages = computed(() => {
  const sid = chatStore.activeSessionId;
  if (!sid) return [];
@@ -131,6 +131,7 @@ export default {
    contextEditSuccess: 'Context length updated',
    contextEditFailed: 'Update failed',
    emptyState: 'Start a conversation with Hermes Agent',
+    cliEmptyState: 'Start a CLI chat session',
    inputPlaceholder: 'Type a message... (Enter to send, Shift+Enter for new line)',
    attachFiles: 'Attach files',
    autoPlaySpeech: 'Auto-play voice',
@@ -159,6 +160,7 @@ export default {
    searchEnterHint: 'Enter to open · Esc to close',
    searchFailed: 'Failed to search sessions',
    newChat: 'New Chat',
+    newCliChat: 'New CLI',
    deleteSession: 'Delete this session?',
    sessionDeleted: 'Session deleted',
    toggleBatchMode: 'Batch selection',
@@ -131,6 +131,7 @@ export default {
    contextEditSuccess: '上下文长度已更新',
    contextEditFailed: '更新失败',
    emptyState: '开始与 Hermes Agent 对话',
+    cliEmptyState: '开始 CLI 对话',
    inputPlaceholder: '输入消息... (Enter 发送，Shift+Enter 换行)',
    attachFiles: '添加附件',
    autoPlaySpeech: '自动播放语音',
@@ -159,6 +160,7 @@ export default {
    searchEnterHint: 'Enter 打开 · Esc 关闭',
    searchFailed: '搜索会话失败',
    newChat: '新建对话',
+    newCliChat: '新建 CLI',
    deleteSession: '确定删除此会话？',
    sessionDeleted: '会话已删除',
    toggleBatchMode: '批量选择',
@@ -1,310 +0,0 @@
-/**
- * Provider registry — single source of truth for both frontend and backend.
- * Synced from hermes-agent hermes_cli/models.py _PROVIDER_MODELS.
- */
-
-export interface ProviderPreset {
-  label: string
-  value: string
-  base_url: string
-  models: string[]
-}
-
-export const PROVIDER_PRESETS: ProviderPreset[] = [
-  {
-    label: 'Anthropic',
-    value: 'anthropic',
-    base_url: 'https://api.anthropic.com',
-    models: [
-      'claude-opus-4-7',
-      'claude-opus-4-6',
-      'claude-sonnet-4-6',
-      'claude-opus-4-5-20251101',
-      'claude-sonnet-4-5-20250929',
-      'claude-opus-4-20250514',
-      'claude-sonnet-4-20250514',
-      'claude-haiku-4-5-20251001',
-    ],
-  },
-  {
-    label: 'Google AI Studio',
-    value: 'gemini',
-    base_url: 'https://generativelanguage.googleapis.com/v1beta/openai',
-    models: [
-      'gemini-3.1-pro-preview',
-      'gemini-3-flash-preview',
-      'gemini-3.1-flash-lite-preview',
-      'gemini-2.5-pro',
-      'gemini-2.5-flash',
-      'gemini-2.5-flash-lite',
-      'gemma-4-31b-it',
-      'gemma-4-26b-it',
-    ],
-  },
-  {
-    label: 'DeepSeek',
-    value: 'deepseek',
-    base_url: 'https://api.deepseek.com',
-    models: ['deepseek-chat', 'deepseek-reasoner'],
-  },
-  {
-    label: 'Z.AI / GLM',
-    value: 'zai',
-    base_url: 'https://api.z.ai/api/paas/v4',
-    models: ['glm-5.1', 'glm-5', 'glm-5v-turbo', 'glm-5-turbo', 'glm-4.7', 'glm-4.5', 'glm-4.5-flash'],
-  },
-  {
-    label: 'Kimi for Coding',
-    value: 'kimi-coding',
-    base_url: 'https://api.kimi.com/coding/v1',
-    models: [
-      'kimi-for-coding',
-      'kimi-k2.5',
-      'kimi-k2-thinking',
-      'kimi-k2-thinking-turbo',
-      'kimi-k2-turbo-preview',
-      'kimi-k2-0905-preview',
-    ],
-  },
-  {
-    label: 'Kimi for Coding (CN)',
-    value: 'kimi-coding-cn',
-    base_url: 'https://api.kimi.com/coding/v1',
-    models: [
-      'kimi-k2.5',
-      'kimi-k2-thinking',
-      'kimi-k2-turbo-preview',
-      'kimi-k2-0905-preview',
-    ],
-  },
-  {
-    label: 'Moonshot',
-    value: 'moonshot',
-    base_url: 'https://api.moonshot.cn/v1',
-    models: [
-      'kimi-k2.5',
-      'kimi-k2-thinking',
-      'kimi-k2-turbo-preview',
-      'kimi-k2-0905-preview',
-    ],
-  },
-  {
-    label: 'xAI',
-    value: 'xai',
-    base_url: 'https://api.x.ai/v1',
-    models: ['grok-4.20-reasoning', 'grok-4-1-fast-reasoning'],
-  },
-  {
-    label: 'MiniMax',
-    value: 'minimax',
-    base_url: 'https://api.minimax.io/anthropic/v1',
-    models: ['MiniMax-M2.7', 'MiniMax-M2.7-highspeed', 'MiniMax-M2.5', 'MiniMax-M2.5-highspeed', 'MiniMax-M2.1', 'MiniMax-M2.1-highspeed', 'MiniMax-M2', 'MiniMax-M2-highspeed'],
-  },
-  {
-    label: 'MiniMax (China)',
-    value: 'minimax-cn',
-    base_url: 'https://api.minimaxi.com/v1',
-    models: ['MiniMax-M2.7', 'MiniMax-M2.7-highspeed', 'MiniMax-M2.5', 'MiniMax-M2.5-highspeed', 'MiniMax-M2.1', 'MiniMax-M2.1-highspeed', 'MiniMax-M2', 'MiniMax-M2-highspeed'],
-  },
-  {
-    label: 'Alibaba Cloud',
-    value: 'alibaba',
-    base_url: 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1',
-    models: [
-      'qwen3.5-plus',
-      'qwen3-coder-plus',
-      'qwen3-coder-next',
-      'glm-5',
-      'glm-4.7',
-      'kimi-k2.5',
-      'MiniMax-M2.5',
-    ],
-  },
-  {
-    label: 'Alibaba Cloud (Coding Plan)',
-    value: 'alibaba-coding-plan',
-    // NOTE: This is the international (intl) DashScope endpoint, matching upstream
-    // hermes-agent (auth.py:255). Mainland China DashScope accounts (sk-sp-* keys
-    // issued by dashscope.aliyun.com) must override via ALIBABA_CODING_PLAN_BASE_URL=
-    // https://coding.dashscope.aliyuncs.com/v1 (no -intl), since the -intl endpoint
-    // returns HTTP 401 for those keys.
-    base_url: 'https://coding-intl.dashscope.aliyuncs.com/v1',
-    models: [
-      'qwen3.5-plus',
-      'qwen3-max-2026-01-23',
-      'qwen3-coder-next',
-      'qwen3-coder-plus',
-      'glm-5',
-      'glm-4.7',
-      'kimi-k2.5',
-      'MiniMax-M2.5',
-    ],
-  },
-  {
-    label: 'Hugging Face',
-    value: 'huggingface',
-    base_url: 'https://router.huggingface.co/v1',
-    models: [
-      'Qwen/Qwen3.5-397B-A17B',
-      'Qwen/Qwen3.5-35B-A3B',
-      'deepseek-ai/DeepSeek-V3.2',
-      'moonshotai/Kimi-K2.5',
-      'MiniMaxAI/MiniMax-M2.5',
-      'zai-org/GLM-5',
-      'XiaomiMiMo/MiMo-V2-Flash',
-      'moonshotai/Kimi-K2-Thinking',
-    ],
-  },
-  {
-    label: 'Xiaomi MiMo',
-    value: 'xiaomi',
-    base_url: 'https://api.xiaomimimo.com/v1',
-    models: ['mimo-v2-pro', 'mimo-v2-omni', 'mimo-v2-flash'],
-  },
-  {
-    label: 'Kilo Code',
-    value: 'kilocode',
-    base_url: 'https://api.kilo.ai/api/gateway',
-    models: [
-      'anthropic/claude-opus-4.6',
-      'anthropic/claude-sonnet-4.6',
-      'openai/gpt-5.4',
-      'google/gemini-3-pro-preview',
-      'google/gemini-3-flash-preview',
-    ],
-  },
-  {
-    label: 'Vercel AI Gateway',
-    value: 'ai-gateway',
-    base_url: 'https://ai-gateway.vercel.sh/v1',
-    models: [
-      'anthropic/claude-opus-4.6',
-      'anthropic/claude-sonnet-4.6',
-      'anthropic/claude-sonnet-4.5',
-      'anthropic/claude-haiku-4.5',
-      'openai/gpt-5',
-      'openai/gpt-4.1',
-      'openai/gpt-4.1-mini',
-      'google/gemini-3-pro-preview',
-      'google/gemini-3-flash',
-      'google/gemini-2.5-pro',
-      'google/gemini-2.5-flash',
-      'deepseek/deepseek-v3.2',
-    ],
-  },
-  {
-    label: 'CLIProxyAPI',
-    value: 'cliproxyapi',
-    base_url: 'http://127.0.0.1:8317/v1',
-    models: [
-      'gpt-5.5',
-      'gpt-5-codex',
-      'claude-sonnet-4-6',
-      'claude-sonnet-4-5-20250929',
-      'gemini-3.1-pro-preview',
-      'gemini-2.5-pro',
-    ],
-  },
-  {
-    label: 'OpenCode Zen',
-    value: 'opencode-zen',
-    base_url: 'https://opencode.ai/zen/v1',
-    models: [
-      'gpt-5.4-pro',
-      'gpt-5.4',
-      'gpt-5.3-codex',
-      'gpt-5.3-codex-spark',
-      'gpt-5.2',
-      'gpt-5.2-codex',
-      'gpt-5.1',
-      'gpt-5.1-codex',
-      'gpt-5.1-codex-max',
-      'gpt-5.1-codex-mini',
-      'gpt-5',
-      'gpt-5-codex',
-      'gpt-5-nano',
-      'claude-opus-4-6',
-      'claude-opus-4-5',
-      'claude-opus-4-1',
-      'claude-sonnet-4-6',
-      'claude-sonnet-4-5',
-      'claude-sonnet-4',
-      'claude-haiku-4-5',
-      'claude-3-5-haiku',
-      'gemini-3.1-pro',
-      'gemini-3-pro',
-      'gemini-3-flash',
-      'minimax-m2.7',
-      'minimax-m2.5',
-      'minimax-m2.5-free',
-      'minimax-m2.1',
-      'glm-5',
-      'glm-4.7',
-      'glm-4.6',
-      'kimi-k2.5',
-      'kimi-k2-thinking',
-      'kimi-k2',
-      'qwen3-coder',
-      'big-pickle',
-    ],
-  },
-  {
-    label: 'OpenCode Go',
-    value: 'opencode-go',
-    base_url: 'https://opencode.ai/zen/go/v1',
-    models: ['glm-5.1', 'glm-5', 'kimi-k2.5', 'mimo-v2-pro', 'mimo-v2-omni', 'minimax-m2.7', 'minimax-m2.5'],
-  },
-  {
-    label: 'OpenAI Codex',
-    value: 'openai-codex',
-    base_url: 'https://chatgpt.com/backend-api/codex',
-    models: ['gpt-5.5', 'gpt-5.4-mini', 'gpt-5.4', 'gpt-5.3-codex', 'gpt-5.2-codex', 'gpt-5.1-codex-max', 'gpt-5.1-codex-mini'],
-  },
-  {
-    label: 'Arcee AI',
-    value: 'arcee',
-    base_url: 'https://api.arcee.ai/v1',
-    models: ['trinity-large-thinking', 'trinity-large-preview', 'trinity-mini'],
-  },
-  {
-    label: 'OpenRouter',
-    value: 'openrouter',
-    base_url: 'https://openrouter.ai/api/v1',
-    models: [],
-  },
-  {
-    label: 'GitHub Copilot',
-    value: 'copilot',
-    base_url: 'https://api.githubcopilot.com',
-    models: [
-      'gpt-5.4',
-      'gpt-5.4-mini',
-      'gpt-5-mini',
-      'gpt-5.3-codex',
-      'gpt-5.2-codex',
-      'gpt-4.1',
-      'gpt-4o',
-      'gpt-4o-mini',
-      'claude-sonnet-4.6',
-      'claude-sonnet-4',
-      'claude-sonnet-4.5',
-      'claude-haiku-4.5',
-      'gemini-3.1-pro-preview',
-      'gemini-3-pro-preview',
-      'gemini-3-flash-preview',
-      'gemini-2.5-pro',
-      'grok-code-fast-1',
-    ],
-  },
-]
-
-/** Build a Record<providerKey, models[]> for backend lookup */
-export function buildProviderModelMap(): Record<string, string[]> {
-  const map: Record<string, string[]> = {}
-  for (const p of PROVIDER_PRESETS) {
-    if (p.models.length > 0) {
-      map[p.value] = p.models
-    }
-  }
-  return map
-}
@@ -1,4 +1,4 @@
-import { startRunViaSocket, resumeSession, registerSessionHandlers, unregisterSessionHandlers, getChatRunSocket, type RunEvent, type ContentBlock as ContentBlockImport } from '@/api/hermes/chat'
+import { startRunViaSocket, resumeSession, registerSessionHandlers, unregisterSessionHandlers, getChatRunSocket, respondToolApproval, type RunEvent, type ContentBlock as ContentBlockImport } from '@/api/hermes/chat'
 import { deleteSession as deleteSessionApi, fetchSession, fetchSessions, type HermesMessage, type SessionSummary } from '@/api/hermes/sessions'
 import { getApiKey } from '@/api/client'
 import { defineStore } from 'pinia'
@@ -43,6 +43,16 @@ export interface Message {
  queued?: boolean
 }

+export interface PendingApproval {
+  sessionId: string
+  approvalId: string
+  command: string
+  description: string
+  choices: Array<'once' | 'session' | 'always' | 'deny'>
+  allowPermanent: boolean
+  requestedAt: number
+}
+
 export interface Session {
  id: string
  title: string
@@ -320,6 +330,11 @@ export const useChatStore = defineStore('chat', () => {
  const queueLengths = ref<Map<string, number>>(new Map())
  /** sessionId → queued user messages not yet visible in the transcript */
  const queuedUserMessages = ref<Map<string, Message[]>>(new Map())
+  const pendingApprovals = ref<Map<string, PendingApproval>>(new Map())
+  const activePendingApproval = computed(() => {
+    const sid = activeSessionId.value
+    return sid ? pendingApprovals.value.get(sid) || null : null
+  })

  // 自动播放语音开关
  const autoPlaySpeechEnabled = ref(false)
@@ -432,6 +447,30 @@ export const useChatStore = defineStore('chat', () => {
    return session
  }

+  function newCliSession(): Session {
+    const now = new Date()
+    const ts = [
+      now.getFullYear(),
+      String(now.getMonth() + 1).padStart(2, '0'),
+      String(now.getDate()).padStart(2, '0'),
+      '_',
+      String(now.getHours()).padStart(2, '0'),
+      String(now.getMinutes()).padStart(2, '0'),
+      String(now.getSeconds()).padStart(2, '0'),
+    ].join('')
+    const hex = Math.random().toString(16).slice(2, 8)
+    const session: Session = {
+      id: `${ts}_${hex}`,
+      title: '',
+      source: 'cli',
+      messages: [],
+      createdAt: Date.now(),
+      updatedAt: Date.now(),
+    }
+    sessions.value.unshift(session)
+    return session
+  }
+
  async function switchSession(sessionId: string, focusId?: string | null) {
    clearThinkingObservationFor(sessionId)
    activeSessionId.value = sessionId
@@ -503,6 +542,49 @@ export const useChatStore = defineStore('chat', () => {
                setAbortState({ aborting: true, synced: null })
              } else if (e.event === 'abort.completed') {
                setAbortState({ aborting: false, synced: e.synced ?? false })
+              } else if (e.event === 'approval.requested') {
+                setPendingApproval({ ...e, session_id: sessionId } as RunEvent)
+              } else if (e.event === 'approval.resolved') {
+                clearPendingApproval({ ...e, session_id: sessionId } as RunEvent)
+              } else if (e.event === 'tool.started') {
+                const msgs = getSessionMsgs(sessionId)
+                const toolCallId = e.tool_call_id as string | undefined
+                const existingTool = toolCallId
+                  ? msgs.find(m => m.role === 'tool' && m.toolCallId === toolCallId)
+                  : null
+                if (existingTool) {
+                  updateMessage(sessionId, existingTool.id, {
+                    toolName: e.tool || e.name,
+                    toolArgs: typeof e.arguments === 'string' ? e.arguments : existingTool.toolArgs,
+                    toolPreview: e.preview || existingTool.toolPreview,
+                    toolStatus: existingTool.toolStatus || 'running',
+                  })
+                } else {
+                  addMessage(sessionId, {
+                    id: uid(),
+                    role: 'tool',
+                    content: '',
+                    timestamp: Date.now(),
+                    toolName: e.tool || e.name,
+                    toolCallId,
+                    toolPreview: e.preview,
+                    toolArgs: typeof e.arguments === 'string' ? e.arguments : undefined,
+                    toolStatus: 'running',
+                  })
+                }
+              } else if (e.event === 'tool.completed') {
+                const msgs = getSessionMsgs(sessionId)
+                const toolCallId = e.tool_call_id as string | undefined
+                const toolMsgs = toolCallId
+                  ? msgs.filter(m => m.role === 'tool' && m.toolCallId === toolCallId)
+                  : msgs.filter(m => m.role === 'tool' && m.toolStatus === 'running')
+                if (toolMsgs.length > 0) {
+                  updateMessage(sessionId, toolMsgs[toolMsgs.length - 1].id, {
+                    toolStatus: e.error === true ? 'error' : 'done',
+                    toolDuration: e.duration,
+                    toolResult: typeof e.output === 'string' ? e.output : undefined,
+                  })
+                }
              }
            }
          }
@@ -603,6 +685,45 @@ export const useChatStore = defineStore('chat', () => {
    })
  }

+  function setPendingApproval(evt: RunEvent) {
+    const sid = evt.session_id
+    const approvalId = (evt as any).approval_id as string | undefined
+    if (!sid || !approvalId) return
+    const rawChoices = Array.isArray((evt as any).choices) ? (evt as any).choices : ['once', 'session', 'deny']
+    const choices = rawChoices
+      .filter((choice: unknown): choice is PendingApproval['choices'][number] =>
+        choice === 'once' || choice === 'session' || choice === 'always' || choice === 'deny')
+    pendingApprovals.value.set(sid, {
+      sessionId: sid,
+      approvalId,
+      command: String((evt as any).command || ''),
+      description: String((evt as any).description || ''),
+      choices: choices.length ? choices : ['once', 'session', 'deny'],
+      allowPermanent: Boolean((evt as any).allow_permanent),
+      requestedAt: Date.now(),
+    })
+    pendingApprovals.value = new Map(pendingApprovals.value)
+  }
+
+  function clearPendingApproval(evt: RunEvent) {
+    const sid = evt.session_id
+    if (!sid) return
+    const current = pendingApprovals.value.get(sid)
+    if (!current) return
+    const approvalId = (evt as any).approval_id
+    if (approvalId && current.approvalId !== approvalId) return
+    pendingApprovals.value.delete(sid)
+    pendingApprovals.value = new Map(pendingApprovals.value)
+  }
+
+  function respondApproval(choice: PendingApproval['choices'][number]) {
+    const pending = activePendingApproval.value
+    if (!pending) return
+    respondToolApproval(pending.sessionId, pending.approvalId, choice)
+    pendingApprovals.value.delete(pending.sessionId)
+    pendingApprovals.value = new Map(pendingApprovals.value)
+  }
+
  function showNextQueuedUserMessage(sessionId: string) {
    const queue = queuedUserMessages.value.get(sessionId)
    if (!queue?.length) return
@@ -715,6 +836,7 @@ export const useChatStore = defineStore('chat', () => {
        session_id: sid,
        model: sessionModel || undefined,
        queue_id: userMsg.id,
+        source: (activeSession.value?.source === 'cli' ? 'cli' : 'api_server') as 'cli' | 'api_server',
      }

      if (shouldQueue) {
@@ -967,6 +1089,16 @@ export const useChatStore = defineStore('chat', () => {
              break
            }

+            case 'approval.requested': {
+              setPendingApproval(evt)
+              break
+            }
+
+            case 'approval.resolved': {
+              clearPendingApproval(evt)
+              break
+            }
+
            case 'run.completed': {
              const msgs = getSessionMsgs(sid)
              const lastMsg = activeAssistantMessageId
@@ -1394,6 +1526,16 @@ export const useChatStore = defineStore('chat', () => {
          break
        }

+        case 'approval.requested': {
+          setPendingApproval(evt)
+          break
+        }
+
+        case 'approval.resolved': {
+          clearPendingApproval(evt)
+          break
+        }
+
        case 'run.completed': {
          const hasQueue = (evt as any).queue_remaining > 0
          if (hasQueue) {
@@ -1689,12 +1831,15 @@ export const useChatStore = defineStore('chat', () => {
    isAborting,
    queueLengths,
    queuedUserMessages,
+    pendingApprovals,
+    activePendingApproval,
    removeQueuedMessage,
    isLoadingSessions,
    sessionsLoaded,
    isLoadingMessages,

    newChat,
+    newCliSession,
    switchSession,
    switchSessionModel,
    addOrUpdateSession,
@@ -1702,6 +1847,7 @@ export const useChatStore = defineStore('chat', () => {
    deleteSession,
    sendMessage,
    stopStreaming,
+    respondApproval,
    loadSessions,
    refreshActiveSession,
    getThinkingObservation,
@@ -240,12 +240,12 @@ watch(hermesSessionsLoaded, (loaded) => {
  if (loaded && hermesSessions.value.length > 0) {
    // Only auto-load if no session is currently active
    if (!historySessionId.value || !hermesSessions.value.find(s => s.id === historySessionId.value)) {
-      // Find first CLI session
+      // Find first CLI session.
      const firstCliSession = hermesSessions.value.find(s => s.source === 'cli')
      if (firstCliSession) {
        // Ensure the CLI group is expanded
-        if (collapsedGroups.value.has('cli')) {
-          collapsedGroups.value = new Set([...collapsedGroups.value].filter(s => s !== 'cli'))
+        if (collapsedGroups.value.has(firstCliSession.source)) {
+          collapsedGroups.value = new Set([...collapsedGroups.value].filter(s => s !== firstCliSession.source))
        }
        // Load session details
        handleSessionClick(firstCliSession.id)
@@ -7,6 +7,26 @@ import { SessionDeleter } from '../../services/hermes/session-deleter'
 import { getGatewayManagerInstance } from '../../services/gateway-bootstrap'
 import { logger } from '../../services/logger'
 import { smartCloneCleanup } from '../../services/hermes/profile-credentials'
+import { detectHermesHome } from '../../services/hermes/hermes-path'
+
+function profileExistsForManualSwitch(name: string): boolean {
+  const base = detectHermesHome()
+  if (!name || name === 'default') return true
+  return existsSync(join(base, 'profiles', name, 'config.yaml')) || existsSync(join(base, 'profiles', name))
+}
+
+async function useProfileWithFallback(name: string): Promise<string> {
+  try {
+    return await hermesCli.useProfile(name)
+  } catch (err: any) {
+    if (!profileExistsForManualSwitch(name)) throw err
+
+    const base = detectHermesHome()
+    writeFileSync(join(base, 'active_profile'), `${name}\n`, 'utf-8')
+    logger.warn(err, '[switchProfile] hermes profile use failed; wrote active_profile directly for existing profile "%s"', name)
+    return `Switched to profile ${name}`
+  }
+}

 export async function list(ctx: any) {
  try {
@@ -159,7 +179,7 @@ export async function switchProfile(ctx: any) {
    return
  }
  try {
-    const output = await hermesCli.useProfile(name)
+    const output = await useProfileWithFallback(name)

    // Verify the active_profile file immediately (Hermes CLI writes synchronously)
    // Quick verification with 2 retries to handle edge cases (filesystem delays, concurrency)
@@ -185,6 +205,16 @@ export async function switchProfile(ctx: any) {
    const mgr = getGatewayManagerInstance()
    if (mgr) { mgr.setActiveProfile(name) }

+    // Destroy all bridge sessions so they get recreated with the new profile config
+    try {
+      const { AgentBridgeClient } = await import('../../services/hermes/agent-bridge')
+      const bridge = new AgentBridgeClient()
+      await bridge.destroyAll()
+      logger.info('[switchProfile] destroyed all bridge sessions for profile "%s"', name)
+    } catch (err: any) {
+      logger.warn(err, '[switchProfile] failed to destroy bridge sessions')
+    }
+
    try {
      const detail = await hermesCli.getProfile(name)
      logger.debug('Profile detail.path = %s', detail.path)
@@ -1,19 +1,16 @@
 import * as hermesCli from '../../services/hermes/hermes-cli'
-import { listConversationSummaries, getConversationDetail } from '../../services/hermes/conversations'
-import { listConversationSummariesFromDb, getConversationDetailFromDb } from '../../db/hermes/conversations-db'
-import { listSessionSummaries, searchSessionSummaries, getUsageStatsFromDb, getSessionDetailFromDb } from '../../db/hermes/sessions-db'
+import { listSessionSummaries, getUsageStatsFromDb, getSessionDetailFromDb } from '../../db/hermes/sessions-db'
 import {
  listSessions as localListSessions,
  searchSessions as localSearchSessions,
  getSessionDetail as localGetSessionDetail,
  deleteSession as localDeleteSession,
  renameSession as localRenameSession,
-  useLocalSessionStore,
 } from '../../db/hermes/session-store'
 import { ExportCompressor } from '../../lib/context-compressor/export-compressor'
 import { getGatewayManagerInstance } from '../../services/gateway-bootstrap'
-import { deleteUsage, getUsage, getUsageBatch, getLocalUsageStats } from '../../db/hermes/usage-store'
-import type { LocalUsageStats, UsageStatsModelRow, UsageStatsDailyRow } from '../../db/hermes/usage-store'
+import { deleteUsage, getUsage, getUsageBatch } from '../../db/hermes/usage-store'
+import type { UsageStatsModelRow, UsageStatsDailyRow } from '../../db/hermes/usage-store'
 import { getModelContextLength } from '../../services/hermes/model-context'
 import { getActiveProfileName } from '../../services/hermes/hermes-profile'
 import { getGroupChatServer } from '../../routes/hermes/group-chat'
@@ -36,130 +33,75 @@ function filterPendingDeletedConversationSummaries(items: ConversationSummary[])

 export async function listConversations(ctx: any) {
  const source = (ctx.query.source as string) || undefined
-  const humanOnly = (ctx.query.humanOnly as string) !== 'false' && ctx.query.humanOnly !== '0'
  const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : undefined

-  if (useLocalSessionStore()) {
-    const profile = getActiveProfileName()
-    const sessions = localListSessions(profile, source, limit && limit > 0 ? limit : 200)
-    const summaries: ConversationSummary[] = sessions.map(s => ({
-      id: s.id,
-      source: s.source,
-      model: s.model,
-      title: s.title,
-      started_at: s.started_at,
-      ended_at: s.ended_at,
-      last_active: s.last_active,
-      message_count: s.message_count,
-      tool_call_count: s.tool_call_count,
-      input_tokens: s.input_tokens,
-      output_tokens: s.output_tokens,
-      cache_read_tokens: s.cache_read_tokens,
-      cache_write_tokens: s.cache_write_tokens,
-      reasoning_tokens: s.reasoning_tokens,
-      billing_provider: s.billing_provider,
-      estimated_cost_usd: s.estimated_cost_usd,
-      actual_cost_usd: s.actual_cost_usd,
-      cost_status: s.cost_status,
-      preview: s.preview,
-      workspace: s.workspace || null,
-      is_active: s.ended_at == null && (Date.now() / 1000 - s.last_active) <= 300,
-      thread_session_count: 1,
-    }))
-    ctx.body = { sessions: filterPendingDeletedConversationSummaries(summaries) }
-    return
-  }
-
-  try {
-    const sessions = await listConversationSummariesFromDb({ source, humanOnly, limit })
-    ctx.body = { sessions: filterPendingDeletedConversationSummaries(sessions) }
-    return
-  } catch (err) {
-    logger.warn(err, 'Hermes Conversation DB: summary query failed, falling back to CLI export')
-  }
-
-  const sessions = await listConversationSummaries({ source, humanOnly, limit })
-  ctx.body = { sessions: filterPendingDeletedConversationSummaries(sessions) }
+  const profile = getActiveProfileName()
+  const sessions = localListSessions(profile, source, limit && limit > 0 ? limit : 200)
+  const summaries: ConversationSummary[] = sessions.map(s => ({
+    id: s.id,
+    source: s.source,
+    model: s.model,
+    title: s.title,
+    started_at: s.started_at,
+    ended_at: s.ended_at,
+    last_active: s.last_active,
+    message_count: s.message_count,
+    tool_call_count: s.tool_call_count,
+    input_tokens: s.input_tokens,
+    output_tokens: s.output_tokens,
+    cache_read_tokens: s.cache_read_tokens,
+    cache_write_tokens: s.cache_write_tokens,
+    reasoning_tokens: s.reasoning_tokens,
+    billing_provider: s.billing_provider,
+    estimated_cost_usd: s.estimated_cost_usd,
+    actual_cost_usd: s.actual_cost_usd,
+    cost_status: s.cost_status,
+    preview: s.preview,
+    workspace: s.workspace || null,
+    is_active: s.ended_at == null && (Date.now() / 1000 - s.last_active) <= 300,
+    thread_session_count: 1,
+  }))
+  ctx.body = { sessions: filterPendingDeletedConversationSummaries(summaries) }
 }

 export async function getConversationMessages(ctx: any) {
-  const source = (ctx.query.source as string) || undefined
  const humanOnly = (ctx.query.humanOnly as string) !== 'false' && ctx.query.humanOnly !== '0'

-  if (useLocalSessionStore()) {
-    const detail = localGetSessionDetail(ctx.params.id)
-    if (!detail) {
-      ctx.status = 404
-      ctx.body = { error: 'Conversation not found' }
-      return
-    }
-    const messages = detail.messages
-      .filter(m => {
-        if (humanOnly && m.role !== 'user' && m.role !== 'assistant') return false
-        if (!m.content) return false
-        return true
-      })
-      .map(m => ({
-        id: m.id,
-        session_id: m.session_id,
-        role: m.role as 'user' | 'assistant',
-        content: m.content,
-        timestamp: m.timestamp,
-      }))
-    ctx.body = {
-      session_id: ctx.params.id,
-      messages,
-      visible_count: messages.length,
-      thread_session_count: 1,
-    }
-    return
-  }
-
-  try {
-    const detail = await getConversationDetailFromDb(ctx.params.id, { source, humanOnly })
-    if (!detail) {
-      ctx.status = 404
-      ctx.body = { error: 'Conversation not found' }
-      return
-    }
-    ctx.body = detail
-    return
-  } catch (err) {
-    logger.warn(err, 'Hermes Conversation DB: detail query failed, falling back to CLI export')
-  }
-
-  const detail = await getConversationDetail(ctx.params.id, { source, humanOnly })
+  const detail = localGetSessionDetail(ctx.params.id)
  if (!detail) {
    ctx.status = 404
    ctx.body = { error: 'Conversation not found' }
    return
  }
-  ctx.body = detail
+  const messages = detail.messages
+    .filter(m => {
+      if (humanOnly && m.role !== 'user' && m.role !== 'assistant') return false
+      if (!m.content) return false
+      return true
+    })
+    .map(m => ({
+      id: m.id,
+      session_id: m.session_id,
+      role: m.role as 'user' | 'assistant',
+      content: m.content,
+      timestamp: m.timestamp,
+    }))
+  ctx.body = {
+    session_id: ctx.params.id,
+    messages,
+    visible_count: messages.length,
+    thread_session_count: 1,
+  }
 }

 export async function list(ctx: any) {
-  if (useLocalSessionStore()) {
-    const source = (ctx.query.source as string) || undefined
-    const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : undefined
-    const profile = getActiveProfileName()
-    const sessions = localListSessions(profile, source, limit && limit > 0 ? limit : 2000)
-    ctx.body = { sessions: filterPendingDeletedSessions(sessions) }
-    return
-  }
-
  const source = (ctx.query.source as string) || undefined
  const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : undefined
+  const profile = getActiveProfileName()
+  const effectiveLimit = limit && limit > 0 ? limit : 2000

-  try {
-    const sessions = await listSessionSummaries(source, limit && limit > 0 ? limit : 2000)
-    ctx.body = { sessions: filterPendingDeletedSessions(sessions) }
-    return
-  } catch (err) {
-    logger.warn(err, 'Hermes Session DB: summary query failed, falling back to CLI')
-  }
-
-  const sessions = await hermesCli.listSessions(source, limit)
-  ctx.body = { sessions: filterPendingDeletedSessions(sessions) }
+  const allSessions = localListSessions(profile, source, effectiveLimit)
+  ctx.body = { sessions: filterPendingDeletedSessions(allSessions.filter(s => s.source === 'api_server' || s.source === 'cli')) }
 }

 /**
@@ -169,58 +111,23 @@ export async function list(ctx: any) {
 export async function listHermesSessions(ctx: any) {
  const source = (ctx.query.source as string) || undefined
  const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : undefined
+  const profile = getActiveProfileName()
+  const effectiveLimit = limit && limit > 0 ? limit : 2000

-  try {
-    const sessions = await listSessionSummaries(source, limit && limit > 0 ? limit : 2000)
-    ctx.body = { sessions: filterPendingDeletedSessions(sessions.filter(s => s.source !== 'api_server')) }
-    return
-  } catch (err) {
-    logger.warn(err, 'Hermes Session DB: summary query failed, falling back to CLI')
-  }
-
-  const sessions = await hermesCli.listSessions(source, limit)
-  ctx.body = { sessions: filterPendingDeletedSessions(sessions.filter(s => s.source !== 'api_server')) }
+  const allSessions = localListSessions(profile, source, effectiveLimit)
+  ctx.body = { sessions: filterPendingDeletedSessions(allSessions.filter(s => s.source !== 'api_server')) }
 }

 export async function search(ctx: any) {
-  if (useLocalSessionStore()) {
-    const q = typeof ctx.query.q === 'string' ? ctx.query.q : ''
-    const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : undefined
-    const profile = getActiveProfileName()
-    const results = localSearchSessions(profile, q, limit && limit > 0 ? limit : 20)
-    ctx.body = { results: filterPendingDeletedSessions(results) }
-    return
-  }
-
  const q = typeof ctx.query.q === 'string' ? ctx.query.q : ''
-  const source = typeof ctx.query.source === 'string' && ctx.query.source.trim()
-    ? ctx.query.source.trim()
-    : undefined
  const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : undefined
-
-  try {
-    const results = await searchSessionSummaries(q, source, limit && limit > 0 ? limit : 20)
-    ctx.body = { results: filterPendingDeletedSessions(results) }
-  } catch (err) {
-    logger.error(err, 'Hermes Session DB: search failed')
-    ctx.status = 500
-    ctx.body = { error: 'Failed to search sessions' }
-  }
+  const profile = getActiveProfileName()
+  const results = localSearchSessions(profile, q, limit && limit > 0 ? limit : 20)
+  ctx.body = { results: filterPendingDeletedSessions(results) }
 }

 export async function get(ctx: any) {
-  if (useLocalSessionStore()) {
-    const session = localGetSessionDetail(ctx.params.id)
-    if (!session) {
-      ctx.status = 404
-      ctx.body = { error: 'Session not found' }
-      return
-    }
-    ctx.body = { session }
-    return
-  }
-
-  const session = await hermesCli.getSession(ctx.params.id)
+  const session = localGetSessionDetail(ctx.params.id)
  if (!session) {
    ctx.status = 404
    ctx.body = { error: 'Session not found' }
@@ -262,21 +169,8 @@ export async function getHermesSession(ctx: any) {
 }

 export async function remove(ctx: any) {
-  if (useLocalSessionStore()) {
-    const sessionId = ctx.params.id
-    const ok = localDeleteSession(sessionId)
-    if (!ok) {
-      ctx.status = 500
-      ctx.body = { error: 'Failed to delete session' }
-      return
-    }
-    deleteUsage(sessionId)
-    ctx.body = { ok: true }
-    return
-  }
-
  const sessionId = ctx.params.id
-  const ok = await hermesCli.deleteSession(sessionId)
+  const ok = localDeleteSession(sessionId)
  if (!ok) {
    ctx.status = 500
    ctx.body = { error: 'Failed to delete session' }
@@ -307,27 +201,14 @@ export async function batchRemove(ctx: any) {
    errors: [] as Array<{ id: string; error: string }>
  }

-  if (useLocalSessionStore()) {
-    for (const id of validIds) {
-      const ok = localDeleteSession(id)
-      if (ok) {
-        deleteUsage(id)
-        results.deleted++
-      } else {
-        results.failed++
-        results.errors.push({ id, error: 'Failed to delete session' })
-      }
-    }
-  } else {
-    for (const id of validIds) {
-      const ok = await hermesCli.deleteSession(id)
-      if (ok) {
-        deleteUsage(id)
-        results.deleted++
-      } else {
-        results.failed++
-        results.errors.push({ id, error: 'Failed to delete session' })
-      }
+  for (const id of validIds) {
+    const ok = localDeleteSession(id)
+    if (ok) {
+      deleteUsage(id)
+      results.deleted++
+    } else {
+      results.failed++
+      results.errors.push({ id, error: 'Failed to delete session' })
    }
  }

@@ -354,30 +235,13 @@ export async function usageSingle(ctx: any) {
 }

 export async function rename(ctx: any) {
-  if (useLocalSessionStore()) {
-    const { title } = ctx.request.body as { title?: string }
-    if (!title || typeof title !== 'string') {
-      ctx.status = 400
-      ctx.body = { error: 'title is required' }
-      return
-    }
-    const ok = localRenameSession(ctx.params.id, title.trim())
-    if (!ok) {
-      ctx.status = 500
-      ctx.body = { error: 'Failed to rename session' }
-      return
-    }
-    ctx.body = { ok: true }
-    return
-  }
-
  const { title } = ctx.request.body as { title?: string }
  if (!title || typeof title !== 'string') {
    ctx.status = 400
    ctx.body = { error: 'title is required' }
    return
  }
-  const ok = await hermesCli.renameSession(ctx.params.id, title.trim())
+  const ok = localRenameSession(ctx.params.id, title.trim())
  if (!ok) {
    ctx.status = 500
    ctx.body = { error: 'Failed to rename session' }
@@ -393,20 +257,14 @@ export async function setWorkspace(ctx: any) {
    ctx.body = { error: 'workspace must be a string or null' }
    return
  }
-  if (useLocalSessionStore()) {
-    const { updateSession, getSession, createSession } = await import('../../db/hermes/session-store')
-    const { getActiveProfileName } = await import('../../services/hermes/hermes-profile')
-    const id = ctx.params.id
-    // Create session if it doesn't exist yet (user may set workspace before sending first message)
-    if (!getSession(id)) {
-      createSession({ id, profile: getActiveProfileName(), title: '' })
-    }
-    updateSession(id, { workspace: workspace || null } as any)
-    ctx.body = { ok: true }
-    return
+  const { updateSession, getSession, createSession } = await import('../../db/hermes/session-store')
+  const { getActiveProfileName } = await import('../../services/hermes/hermes-profile')
+  const id = ctx.params.id
+  if (!getSession(id)) {
+    createSession({ id, profile: getActiveProfileName(), title: '' })
  }
-  ctx.status = 501
-  ctx.body = { error: 'Workspace setting only supported in local session store mode' }
+  updateSession(id, { workspace: workspace || null } as any)
+  ctx.body = { ok: true }
 }

 export async function contextLength(ctx: any) {
@@ -418,11 +276,6 @@ export async function usageStats(ctx: any) {
  const rawDays = parseInt(String(ctx.query?.days ?? '30'), 10)
  const days = Number.isFinite(rawDays) && rawDays > 0 ? Math.min(rawDays, 365) : 30

-  // Local Web UI chat usage is kept in the dashboard DB and must be merged
-  // with Hermes' native state.db analytics for the same period.
-  const currentProfile = getActiveProfileName()
-  const local = getLocalUsageStats(currentProfile, days)
-
  let hermes = {
    input_tokens: 0,
    output_tokens: 0,
@@ -442,29 +295,6 @@ export async function usageStats(ctx: any) {
    logger.warn(err, 'usageStats: failed to load Hermes usage analytics from state.db')
  }

-  const totalInput = local.input_tokens + hermes.input_tokens
-  const totalOutput = local.output_tokens + hermes.output_tokens
-  const totalCacheRead = local.cache_read_tokens + hermes.cache_read_tokens
-  const totalCacheWrite = local.cache_write_tokens + hermes.cache_write_tokens
-  const totalReasoning = local.reasoning_tokens + hermes.reasoning_tokens
-  const totalSessions = local.sessions + hermes.sessions
-
-  const modelMap = new Map<string, UsageStatsModelRow>()
-  for (const m of [...local.by_model, ...hermes.by_model]) {
-    const model = (m.model || '').trim() || 'unknown'
-    const existing = modelMap.get(model)
-    if (existing) {
-      existing.input_tokens += m.input_tokens
-      existing.output_tokens += m.output_tokens
-      existing.cache_read_tokens += m.cache_read_tokens
-      existing.cache_write_tokens += m.cache_write_tokens
-      existing.reasoning_tokens += m.reasoning_tokens
-      existing.sessions += m.sessions
-    } else {
-      modelMap.set(model, { ...m, model })
-    }
-  }
-
  const dayMap = new Map<string, UsageStatsDailyRow>()
  const now = new Date()
  for (let i = days - 1; i >= 0; i--) {
@@ -473,7 +303,7 @@ export async function usageStats(ctx: any) {
    const key = d.toISOString().slice(0, 10)
    dayMap.set(key, { date: key, input_tokens: 0, output_tokens: 0, cache_read_tokens: 0, cache_write_tokens: 0, sessions: 0, errors: 0, cost: 0 })
  }
-  for (const d of [...local.by_day, ...hermes.by_day]) {
+  for (const d of hermes.by_day) {
    const existing = dayMap.get(d.date)
    if (existing) {
      existing.input_tokens += d.input_tokens; existing.output_tokens += d.output_tokens
@@ -483,16 +313,16 @@ export async function usageStats(ctx: any) {
  }

  ctx.body = {
-    total_input_tokens: totalInput,
-    total_output_tokens: totalOutput,
-    total_cache_read_tokens: totalCacheRead,
-    total_cache_write_tokens: totalCacheWrite,
-    total_reasoning_tokens: totalReasoning,
-    total_sessions: totalSessions,
+    total_input_tokens: hermes.input_tokens,
+    total_output_tokens: hermes.output_tokens,
+    total_cache_read_tokens: hermes.cache_read_tokens,
+    total_cache_write_tokens: hermes.cache_write_tokens,
+    total_reasoning_tokens: hermes.reasoning_tokens,
+    total_sessions: hermes.sessions,
    total_cost: hermes.cost,
    total_api_calls: hermes.total_api_calls,
    period_days: days,
-    model_usage: [...modelMap.values()].sort((a, b) => (b.input_tokens + b.output_tokens) - (a.input_tokens + a.output_tokens)),
+    model_usage: hermes.by_model.sort((a, b) => (b.input_tokens + b.output_tokens) - (a.input_tokens + a.output_tokens)),
    daily_usage: [...dayMap.values()],
  }
 }
@@ -545,20 +375,7 @@ export async function listWorkspaceFolders(ctx: any) {
 const exportCompressor = new ExportCompressor()

 export async function exportSession(ctx: any) {
-  let session: any = null
-
-  if (useLocalSessionStore()) {
-    session = localGetSessionDetail(ctx.params.id)
-  } else {
-    try {
-      session = await getSessionDetailFromDb(ctx.params.id)
-    } catch (err) {
-      logger.warn(err, 'Hermes Session DB: export detail query failed, falling back to CLI')
-    }
-    if (!session) {
-      session = await hermesCli.getSession(ctx.params.id)
-    }
-  }
+  const session = localGetSessionDetail(ctx.params.id)

  if (!session) {
    ctx.status = 404
@@ -630,38 +447,32 @@ export async function getConversationMessagesPaginated(ctx: any) {
  const offset = ctx.query.offset ? parseInt(ctx.query.offset as string, 10) : 0
  const limit = ctx.query.limit ? parseInt(ctx.query.limit as string, 10) : 50

-  if (useLocalSessionStore()) {
-    const { getSessionDetailPaginated } = await import('../../db/hermes/session-store')
-    const result = getSessionDetailPaginated(ctx.params.id, offset, limit)
+  const { getSessionDetailPaginated } = await import('../../db/hermes/session-store')
+  const result = getSessionDetailPaginated(ctx.params.id, offset, limit)

-    if (!result) {
-      ctx.status = 404
-      ctx.body = { error: 'Conversation not found' }
-      return
-    }
-
-    ctx.body = {
-      session: {
-        id: result.session.id,
-        source: result.session.source,
-        model: result.session.model,
-        title: result.session.title,
-        started_at: result.session.started_at,
-        ended_at: result.session.ended_at,
-        last_active: result.session.last_active,
-        message_count: result.session.message_count,
-        input_tokens: result.session.input_tokens,
-        output_tokens: result.session.output_tokens,
-      },
-      messages: result.messages,
-      total: result.total,
-      offset: result.offset,
-      limit: result.limit,
-      hasMore: result.hasMore,
-    }
+  if (!result) {
+    ctx.status = 404
+    ctx.body = { error: 'Conversation not found' }
    return
  }

-  ctx.status = 404
-  ctx.body = { error: 'Conversation not found' }
+  ctx.body = {
+    session: {
+      id: result.session.id,
+      source: result.session.source,
+      model: result.session.model,
+      title: result.session.title,
+      started_at: result.session.started_at,
+      ended_at: result.session.ended_at,
+      last_active: result.session.last_active,
+      message_count: result.session.message_count,
+      input_tokens: result.session.input_tokens,
+      output_tokens: result.session.output_tokens,
+    },
+    messages: result.messages,
+    total: result.total,
+    offset: result.offset,
+    limit: result.limit,
+    hasMore: result.hasMore,
+  }
 }
@@ -129,14 +129,16 @@ function mapMessageRow(row: Record<string, unknown>): HermesMessageRow {
 export function createSession(data: {
  id: string
  profile?: string
+  source?: string
  model?: string
  title?: string
  workspace?: string
 }): HermesSessionRow {
  const now = Math.floor(Date.now() / 1000)
+  const source = data.source || 'api_server'
  if (!isSqliteAvailable()) {
    return {
-      id: data.id, profile: data.profile || 'default', source: 'api_server',
+      id: data.id, profile: data.profile || 'default', source,
      user_id: null, model: data.model || '', title: data.title || null,
      started_at: now, ended_at: null, end_reason: null,
      message_count: 0, tool_call_count: 0,
@@ -148,8 +150,8 @@ export function createSession(data: {
  const db = getDb()!
  db.prepare(
    `INSERT INTO ${SESSIONS_TABLE} (id, profile, source, model, title, started_at, last_active, workspace)
-     VALUES (?, ?, 'api_server', ?, ?, ?, ?, ?)`,
-  ).run(data.id, data.profile || 'default', data.model || '', data.title || null, now, now, data.workspace || null)
+     VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,
+  ).run(data.id, data.profile || 'default', source, data.model || '', data.title || null, now, now, data.workspace || null)
  return getSession(data.id)!
 }

@@ -565,6 +565,10 @@ function aggregateSessionDetail(
  }
 }

+function chainOrderSql(ids: string[]): string {
+  return ids.map((_, index) => `WHEN ? THEN ${index}`).join(' ')
+}
+
 async function openSessionDb() {
  if (!SQLITE_AVAILABLE) {
    throw new Error(`node:sqlite requires Node >= 22.5, current: ${process.versions.node}`)
@@ -598,7 +602,7 @@ export async function getSessionMessagesFromDb(sessionId: string): Promise<{
    const messageRows = db.prepare(`
      SELECT * FROM messages
      WHERE session_id = ?
-      ORDER BY timestamp, id
+      ORDER BY id
    `).all(sessionId) as Record<string, unknown>[]

    return {
@@ -622,11 +626,12 @@ export async function getSessionDetailFromDb(sessionId: string): Promise<HermesS

    const ids = chain.map(session => session.id)
    const placeholders = ids.map(() => '?').join(', ')
+    const orderSql = chainOrderSql(ids)
    const messageRows = db.prepare(`
      SELECT * FROM messages
      WHERE session_id IN (${placeholders})
-      ORDER BY timestamp, id
-    `).all(...ids) as Record<string, unknown>[]
+      ORDER BY CASE session_id ${orderSql} ELSE ${ids.length} END, id
+    `).all(...ids, ...ids) as Record<string, unknown>[]
    const messages = messageRows.map(mapMessageRow)
    return aggregateSessionDetail(chain, messages, sessionId)
  } finally {
@@ -648,11 +653,12 @@ export async function getSessionDetailFromDbWithProfile(sessionId: string, profi

    const ids = chain.map(session => session.id)
    const placeholders = ids.map(() => '?').join(', ')
+    const orderSql = chainOrderSql(ids)
    const messageRows = db.prepare(`
      SELECT * FROM messages
      WHERE session_id IN (${placeholders})
-      ORDER BY timestamp, id
-    `).all(...ids) as Record<string, unknown>[]
+      ORDER BY CASE session_id ${orderSql} ELSE ${ids.length} END, id
+    `).all(...ids, ...ids) as Record<string, unknown>[]
    const messages = messageRows.map(mapMessageRow)
    return aggregateSessionDetail(chain, messages, sessionId)
  } finally {
@@ -672,7 +678,7 @@ export async function getExactSessionDetailFromDbWithProfile(sessionId: string,
    const messageRows = db.prepare(`
      SELECT * FROM messages
      WHERE session_id = ?
-      ORDER BY timestamp, id
+      ORDER BY id
    `).all(sessionId) as Record<string, unknown>[]
    const messages = messageRows.map(mapMessageRow)
    return aggregateSessionDetail([requested], messages, sessionId)
@@ -818,10 +824,6 @@ export async function getUsageStatsFromDb(
    const apiCallsExpr = tableHasColumn(db, 'sessions', 'api_call_count')
      ? 'COALESCE(SUM(api_call_count), 0)'
      : '0'
-    const sourceFilter = tableHasColumn(db, 'sessions', 'source')
-      ? " AND COALESCE(source, '') != 'api_server'"
-      : ''
-
    const totals = db.prepare(`
      SELECT
        COALESCE(SUM(input_tokens), 0) AS input_tokens,
@@ -833,7 +835,7 @@ export async function getUsageStatsFromDb(
        COUNT(*) AS sessions,
        ${apiCallsExpr} AS total_api_calls
      FROM sessions
-      WHERE started_at > ?${sourceFilter}
+      WHERE started_at > ?
    `).get(since) as Record<string, unknown> | undefined

    if (!totals) return empty
@@ -848,7 +850,7 @@ export async function getUsageStatsFromDb(
        COALESCE(SUM(reasoning_tokens), 0) AS reasoning_tokens,
        COUNT(*) AS sessions
      FROM sessions
-      WHERE started_at > ?${sourceFilter} AND model IS NOT NULL
+      WHERE started_at > ? AND model IS NOT NULL
      GROUP BY model
      ORDER BY COALESCE(SUM(input_tokens), 0) + COALESCE(SUM(output_tokens), 0) DESC
    `).all(since).map(row => ({
@@ -871,7 +873,7 @@ export async function getUsageStatsFromDb(
        COUNT(*) AS sessions,
        COALESCE(SUM(COALESCE(actual_cost_usd, estimated_cost_usd, 0)), 0) AS cost
      FROM sessions
-      WHERE started_at > ?${sourceFilter}
+      WHERE started_at > ?
      GROUP BY date
      ORDER BY date ASC
    `).all(since).map(row => ({
@@ -20,6 +20,7 @@ import { setGroupChatServer } from './routes/hermes/group-chat'
 import { setChatRunServer } from './routes/hermes/chat-run'
 import { GroupChatServer } from './services/hermes/group-chat'
 import { ChatRunSocket } from './services/hermes/chat-run-socket'
+import { startAgentBridgeManager } from './services/hermes/agent-bridge'
 import { logger } from './services/logger'

 // Injected by esbuild at build time; fallback to reading package.json in dev mode
@@ -46,6 +47,7 @@ process.on('unhandledRejection', (reason) => {
 let server: any = null
 let servers: any[] = []
 let chatRunServer: any = null
+let agentBridgeManager: any = null

 interface ListenResult {
  primary: any
@@ -94,6 +96,13 @@ export async function bootstrap() {

  await initGatewayManager()
  console.log('[bootstrap] gateway manager initialized')
+  try {
+    agentBridgeManager = await startAgentBridgeManager()
+    console.log('[bootstrap] agent bridge started')
+  } catch (err) {
+    logger.warn(err, '[bootstrap] agent bridge failed to start')
+    console.warn('[bootstrap] agent bridge failed to start:', err instanceof Error ? err.message : err)
+  }
  await new Promise(resolve => setTimeout(resolve, 1000))
  // Initialize all web-ui SQLite tables
  const { initAllStores } = await import('./db/hermes/init')
@@ -102,11 +111,6 @@ export async function bootstrap() {
  await new Promise(resolve => setTimeout(resolve, 1000))
  console.log('[bootstrap] all stores initialized')

-  // Sync Hermes sessions from all profiles (only if local DB is empty)
-  const { syncAllHermesSessionsOnStartup } = await import('./services/hermes/session-sync')
-  await syncAllHermesSessionsOnStartup()
-  console.log('[bootstrap] Hermes session sync completed')
-
  app.use(cors({ origin: config.corsOrigins }))
  app.use(bodyParser())
  console.log('[bootstrap] cors + bodyParser registered')
@@ -187,7 +191,7 @@ export async function bootstrap() {
    })
  })

-  bindShutdown(servers, groupChatServer, chatRunServer)
+  bindShutdown(servers, groupChatServer, chatRunServer, agentBridgeManager)
  startVersionCheck()
 }

@@ -0,0 +1,85 @@
+# Agent Bridge
+
+Optional backend-side bridge for talking to `~/.hermes/hermes-agent` by
+instantiating `run_agent.AIAgent` directly in a Python process.
+
+This is intentionally separate from the current Web UI chat path.
+
+## Python Service
+
+```bash
+python packages/server/src/services/hermes/agent-bridge/hermes_bridge.py
+```
+
+Default endpoint:
+
+```text
+ipc:///tmp/hermes-agent-bridge.sock
+```
+
+On Windows, the default endpoint is TCP because Python may not support Unix
+domain sockets there:
+
+```text
+tcp://127.0.0.1:18765
+```
+
+Override with:
+
+```bash
+HERMES_AGENT_BRIDGE_ENDPOINT=tcp://127.0.0.1:8765 python packages/server/src/services/hermes/agent-bridge/hermes_bridge.py
+```
+
+The service discovers Hermes Agent in this order:
+
+1. `--agent-root`
+2. `HERMES_AGENT_ROOT`
+3. the installed `hermes` command path
+4. current working directory and parent directories
+5. common locations such as `~/.hermes/hermes-agent`, `~/hermes-agent`, and `/opt/hermes-agent`
+
+Hermes home is resolved from `--hermes-home`, `HERMES_HOME`, then `~/.hermes`.
+
+Default agent root:
+
+```text
+~/.hermes/hermes-agent
+```
+
+You can pass both paths explicitly:
+
+```bash
+python packages/server/src/services/hermes/agent-bridge/hermes_bridge.py \
+  --agent-root ~/.hermes/hermes-agent \
+  --hermes-home ~/.hermes
+```
+
+The socket transport uses Python and Node standard libraries. No ZMQ dependency
+is required.
+
+## Backend Usage
+
+```ts
+import { AgentBridgeClient } from './services/hermes/agent-bridge'
+
+const bridge = new AgentBridgeClient()
+const run = await bridge.chat(sessionId, message)
+
+for await (const chunk of bridge.streamOutput(run.run_id)) {
+  if (chunk.delta) {
+    // forward chunk.delta to Socket.IO/SSE/etc.
+  }
+}
+```
+
+The external chat call only sends `session_id` and `message`. Provider, model,
+keys, tools, reasoning, and session DB are resolved by hermes-agent from the
+normal Hermes config and environment.
+
+The bridge instantiates `AIAgent` with `platform="cli"` by default so behavior
+matches CLI chat. Override it only if a caller intentionally needs a distinct
+platform identity:
+
+```bash
+HERMES_AGENT_BRIDGE_PLATFORM=agent-bridge python packages/server/src/services/hermes/agent-bridge/hermes_bridge.py
+```
@@ -0,0 +1,330 @@
+import { setTimeout as delay } from 'timers/promises'
+import { createConnection, type Socket } from 'net'
+import { URL } from 'url'
+
+export const DEFAULT_AGENT_BRIDGE_ENDPOINT = process.platform === 'win32'
+  ? 'tcp://127.0.0.1:18765'
+  : 'ipc:///tmp/hermes-agent-bridge.sock'
+export const DEFAULT_AGENT_BRIDGE_TIMEOUT_MS = 120000
+
+function envPositiveInt(name: string): number | undefined {
+  const raw = process.env[name]
+  if (!raw) return undefined
+  const value = Number(raw)
+  return Number.isFinite(value) && value > 0 ? value : undefined
+}
+
+export type AgentBridgeStatus = 'running' | 'complete' | 'interrupted' | 'error'
+
+export interface AgentBridgeOptions {
+  endpoint?: string
+  timeoutMs?: number
+}
+
+export interface AgentBridgeRequestOptions {
+  timeoutMs?: number
+}
+
+export type AgentBridgeMessage =
+  | string
+  | Array<Record<string, unknown>>
+
+export interface AgentBridgeResponse {
+  ok: true
+  [key: string]: unknown
+}
+
+export interface AgentBridgeChatStarted extends AgentBridgeResponse {
+  run_id: string
+  session_id: string
+  status: AgentBridgeStatus
+}
+
+export interface AgentBridgeOutput extends AgentBridgeResponse {
+  run_id: string
+  session_id: string
+  status: AgentBridgeStatus
+  delta: string
+  cursor: number
+  output: string
+  done: boolean
+  result?: unknown
+  error?: string | null
+  events: Array<Record<string, unknown>>
+  event_cursor: number
+}
+
+export interface AgentBridgeRunResult extends AgentBridgeResponse {
+  run_id: string
+  session_id: string
+  status: AgentBridgeStatus
+  output: string
+  deltas: string[]
+  events: unknown[]
+  result?: unknown
+  error?: string | null
+}
+
+export interface AgentBridgeCommandResult extends AgentBridgeResponse {
+  session_id: string
+  command: string
+  handled: boolean
+  message?: string
+  new_session_id?: string
+  history?: unknown[]
+  retry?: boolean
+  retry_input?: AgentBridgeMessage
+  title?: string
+}
+
+export class AgentBridgeError extends Error {
+  response?: unknown
+}
+
+export class AgentBridgeClient {
+  readonly endpoint: string
+  readonly timeoutMs: number
+  private lock: Promise<unknown> = Promise.resolve()
+
+  constructor(options: AgentBridgeOptions = {}) {
+    this.endpoint = options.endpoint || process.env.HERMES_AGENT_BRIDGE_ENDPOINT || DEFAULT_AGENT_BRIDGE_ENDPOINT
+    this.timeoutMs = options.timeoutMs ?? envPositiveInt('HERMES_AGENT_BRIDGE_TIMEOUT_MS') ?? DEFAULT_AGENT_BRIDGE_TIMEOUT_MS
+  }
+
+  async connect(): Promise<this> {
+    return this
+  }
+
+  async close(): Promise<void> {
+    return undefined
+  }
+
+  private connectSocket(): Promise<Socket> {
+    return new Promise((resolveConnect, rejectConnect) => {
+      const endpoint = this.endpoint
+      let socket: Socket
+      if (endpoint.startsWith('ipc://')) {
+        socket = createConnection(endpoint.slice('ipc://'.length))
+      } else if (endpoint.startsWith('tcp://')) {
+        const url = new URL(endpoint)
+        socket = createConnection({
+          host: url.hostname || '127.0.0.1',
+          port: Number(url.port),
+        })
+      } else {
+        rejectConnect(new Error(`unsupported agent bridge endpoint: ${endpoint}`))
+        return
+      }
+
+      const cleanup = () => {
+        socket.off('connect', onConnect)
+        socket.off('error', onError)
+      }
+      const onConnect = () => {
+        cleanup()
+        resolveConnect(socket)
+      }
+      const onError = (err: Error) => {
+        cleanup()
+        socket.destroy()
+        rejectConnect(err)
+      }
+      socket.once('connect', onConnect)
+      socket.once('error', onError)
+    })
+  }
+
+  private readResponse(socket: Socket, timeoutMs: number): Promise<string> {
+    return new Promise((resolveRead, rejectRead) => {
+      let buffer = ''
+      const timeout = timeoutMs > 0
+        ? setTimeout(() => {
+            cleanup()
+            socket.destroy()
+            rejectRead(new Error(`Agent bridge request timed out after ${timeoutMs}ms`))
+          }, timeoutMs)
+        : null
+
+      const cleanup = () => {
+        if (timeout) clearTimeout(timeout)
+        socket.off('data', onData)
+        socket.off('error', onError)
+        socket.off('end', onEnd)
+        socket.off('close', onClose)
+      }
+      const finish = (line: string) => {
+        cleanup()
+        socket.end()
+        resolveRead(line)
+      }
+      const onData = (chunk: Buffer) => {
+        buffer += chunk.toString('utf8')
+        const idx = buffer.indexOf('\n')
+        if (idx >= 0) finish(buffer.slice(0, idx))
+      }
+      const onError = (err: Error) => {
+        cleanup()
+        socket.destroy()
+        rejectRead(err)
+      }
+      const onEnd = () => {
+        const line = buffer.trim()
+        if (line) finish(line)
+      }
+      const onClose = () => {
+        if (!buffer.trim()) {
+          cleanup()
+          rejectRead(new Error('Agent bridge socket closed without a response'))
+        }
+      }
+
+      socket.on('data', onData)
+      socket.once('error', onError)
+      socket.once('end', onEnd)
+      socket.once('close', onClose)
+    })
+  }
+
+  async request<T extends AgentBridgeResponse = AgentBridgeResponse>(
+    payload: Record<string, unknown>,
+    options: AgentBridgeRequestOptions = {},
+  ): Promise<T> {
+    const run = async (): Promise<T> => {
+      const timeoutMs = options.timeoutMs || this.timeoutMs
+      const socket = await this.connectSocket()
+      socket.write(`${JSON.stringify(payload)}\n`)
+      const raw = await this.readResponse(socket, timeoutMs)
+      const response = JSON.parse(raw) as { ok?: boolean; error?: string }
+      if (!response.ok) {
+        const error = new AgentBridgeError(response.error || 'Agent bridge request failed')
+        error.response = response
+        throw error
+      }
+      return response as T
+    }
+
+    const next = this.lock.then(run, run)
+    this.lock = next.catch(() => undefined)
+    return next
+  }
+
+  ping(): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'ping' })
+  }
+
+  chat(
+    sessionId: string,
+    message: AgentBridgeMessage,
+    conversationHistory?: unknown[],
+    instructions?: string,
+    profile?: string,
+  ): Promise<AgentBridgeChatStarted> {
+    return this.request<AgentBridgeChatStarted>({
+      action: 'chat',
+      session_id: sessionId,
+      message,
+      ...(conversationHistory ? { conversation_history: conversationHistory } : {}),
+      ...(instructions ? { instructions } : {}),
+      ...(profile ? { profile } : {}),
+    })
+  }
+
+  command(sessionId: string, command: string): Promise<AgentBridgeCommandResult> {
+    return this.request<AgentBridgeCommandResult>({
+      action: 'command',
+      session_id: sessionId,
+      command,
+    })
+  }
+
+  getOutput(runId: string, cursor = 0, eventCursor = 0, options: AgentBridgeRequestOptions = {}): Promise<AgentBridgeOutput> {
+    return this.request<AgentBridgeOutput>({
+      action: 'get_output',
+      run_id: runId,
+      cursor,
+      event_cursor: eventCursor,
+    }, options)
+  }
+
+  async *streamOutput(
+    runId: string,
+    options: AgentBridgeRequestOptions & { intervalMs?: number } = {},
+  ): AsyncGenerator<AgentBridgeOutput> {
+    const intervalMs = options.intervalMs || 100
+    let cursor = 0
+    let eventCursor = 0
+    for (;;) {
+      const chunk = await this.getOutput(runId, cursor, eventCursor, options)
+      cursor = chunk.cursor
+      eventCursor = chunk.event_cursor
+      if (chunk.delta || chunk.done || (chunk.events && chunk.events.length > 0)) yield chunk
+      if (chunk.done) return
+      await delay(intervalMs)
+    }
+  }
+
+  async chatStream(
+    sessionId: string,
+    message: AgentBridgeMessage,
+    onDelta: (delta: string, chunk: AgentBridgeOutput) => void | Promise<void>,
+    options: AgentBridgeRequestOptions & { intervalMs?: number } = {},
+  ): Promise<AgentBridgeOutput> {
+    const started = await this.chat(sessionId, message)
+    let last: AgentBridgeOutput | null = null
+    for await (const chunk of this.streamOutput(started.run_id, options)) {
+      last = chunk
+      if (chunk.delta) await onDelta(chunk.delta, chunk)
+    }
+    if (!last) throw new Error(`Agent bridge run ${started.run_id} produced no output state`)
+    return last
+  }
+
+  getResult(runId: string, options: AgentBridgeRequestOptions = {}): Promise<AgentBridgeRunResult> {
+    return this.request<AgentBridgeRunResult>({ action: 'get_result', run_id: runId }, options)
+  }
+
+  interrupt(sessionId: string, message?: string): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'interrupt', session_id: sessionId, message })
+  }
+
+  steer(sessionId: string, text: string): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'steer', session_id: sessionId, text })
+  }
+
+  approvalRespond(approvalId: string, choice: string): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'approval_respond', approval_id: approvalId, choice })
+  }
+
+  compressionRespond(
+    requestId: string,
+    payload: { messages?: unknown[]; system_message?: string; error?: string },
+  ): Promise<AgentBridgeResponse> {
+    return this.request({
+      action: 'compression_respond',
+      request_id: requestId,
+      ...payload,
+    }, { timeoutMs: this.timeoutMs })
+  }
+
+  destroyAll(): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'destroy_all' })
+  }
+
+  getHistory(sessionId: string): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'get_history', session_id: sessionId })
+  }
+
+  destroy(sessionId: string): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'destroy', session_id: sessionId })
+  }
+
+  list(): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'list' })
+  }
+
+  shutdown(): Promise<AgentBridgeResponse> {
+    return this.request({ action: 'shutdown' })
+  }
+}
+
+export default AgentBridgeClient
@@ -0,0 +1,2 @@
+export * from './client'
+export * from './manager'
@@ -0,0 +1,360 @@
+import { execFileSync, spawn, type ChildProcess } from 'child_process'
+import { existsSync, readFileSync } from 'fs'
+import { dirname, isAbsolute, join, resolve } from 'path'
+import { logger } from '../../logger'
+import { detectHermesHome, getHermesBin } from '../hermes-path'
+import { DEFAULT_AGENT_BRIDGE_ENDPOINT } from './client'
+
+const DEFAULT_AGENT_BRIDGE_STARTUP_TIMEOUT_MS = 120000
+
+export interface AgentBridgeManagerOptions {
+  endpoint?: string
+  python?: string
+  agentRoot?: string
+  hermesHome?: string
+  startupTimeoutMs?: number
+}
+
+interface BridgeCommand {
+  command: string
+  argsPrefix: string[]
+  agentRoot?: string
+  hermesHome: string
+}
+
+function envPositiveInt(name: string): number | undefined {
+  const raw = process.env[name]
+  if (!raw) return undefined
+  const value = Number(raw)
+  return Number.isFinite(value) && value > 0 ? value : undefined
+}
+
+function pathCandidates(agentRoot?: string): string[] {
+  if (!agentRoot) return []
+  return process.platform === 'win32'
+    ? [
+        join(agentRoot, 'venv', 'Scripts', 'python.exe'),
+        join(agentRoot, 'venv', 'Scripts', 'python3.exe'),
+        join(agentRoot, '.venv', 'Scripts', 'python.exe'),
+        join(agentRoot, '.venv', 'Scripts', 'python3.exe'),
+      ]
+    : [
+        join(agentRoot, 'venv', 'bin', 'python3'),
+        join(agentRoot, 'venv', 'bin', 'python'),
+        join(agentRoot, '.venv', 'bin', 'python3'),
+        join(agentRoot, '.venv', 'bin', 'python'),
+      ]
+}
+
+function uvCandidates(agentRoot?: string): string[] {
+  return [
+    process.env.HERMES_AGENT_BRIDGE_UV,
+    process.env.UV,
+    ...(process.platform === 'win32'
+      ? [
+          agentRoot ? join(agentRoot, 'venv', 'Scripts', 'uv.exe') : '',
+          agentRoot ? join(agentRoot, 'venv', 'Scripts', 'uv.cmd') : '',
+          agentRoot ? join(agentRoot, '.venv', 'Scripts', 'uv.exe') : '',
+          agentRoot ? join(agentRoot, '.venv', 'Scripts', 'uv.cmd') : '',
+        ]
+      : [
+          agentRoot ? join(agentRoot, 'venv', 'bin', 'uv') : '',
+          agentRoot ? join(agentRoot, '.venv', 'bin', 'uv') : '',
+        ]),
+    'uv',
+  ].filter((value): value is string => !!value && value.trim().length > 0)
+}
+
+function resolveExecutable(command: string): string | undefined {
+  const trimmed = command.trim()
+  if (!trimmed) return undefined
+  if (isAbsolute(trimmed) || trimmed.includes('/') || trimmed.includes('\\')) {
+    return existsSync(trimmed) ? resolve(trimmed) : undefined
+  }
+  try {
+    const lookup = process.platform === 'win32'
+      ? execFileSync('where.exe', [trimmed], { encoding: 'utf-8', windowsHide: true })
+      : execFileSync('which', [trimmed], { encoding: 'utf-8' })
+    return lookup.split(/\r?\n/).map(line => line.trim()).find(Boolean)
+  } catch {
+    return undefined
+  }
+}
+
+function agentRootFromHermesBin(): string | undefined {
+  const hermesBin = resolveExecutable(getHermesBin())
+  if (!hermesBin) return undefined
+
+  const binDir = dirname(hermesBin)
+  const rootCandidates = [
+    resolve(binDir, '..'),
+    resolve(binDir, '..', '..'),
+    resolve(binDir, '..', 'hermes-agent'),
+    resolve(binDir, '..', '..', 'hermes-agent'),
+  ]
+  const root = rootCandidates.find(candidate => existsSync(join(candidate, 'run_agent.py')))
+  if (root) return root
+
+  try {
+    const first = readFileSync(hermesBin, 'utf-8').split(/\r?\n/, 1)[0]
+    const match = first.match(/^#!\s*(.+)$/)
+    const python = match?.[1]?.trim().split(/\s+/)[0]
+    if (python) {
+      const pyDir = dirname(python)
+      const shebangRootCandidates = [
+        resolve(pyDir, '..', '..'),
+        resolve(pyDir, '..', '..', 'hermes-agent'),
+      ]
+      return shebangRootCandidates.find(candidate => existsSync(join(candidate, 'run_agent.py')))
+    }
+  } catch {}
+  return undefined
+}
+
+function hermesBinPython(): string | undefined {
+  const hermesBin = resolveExecutable(getHermesBin())
+  if (!hermesBin) return undefined
+  try {
+    const first = readFileSync(hermesBin, 'utf-8').split(/\r?\n/, 1)[0]
+    const match = first.match(/^#!\s*(.+)$/)
+    const python = match?.[1]?.trim().split(/\s+/)[0]
+    return python && existsSync(python) ? python : undefined
+  } catch {
+    return undefined
+  }
+}
+
+function firstExistingExecutable(candidates: string[]): string | undefined {
+  for (const candidate of candidates) {
+    if (!isAbsolute(candidate) && !candidate.includes('/') && !candidate.includes('\\')) {
+      const resolved = resolveExecutable(candidate)
+      if (resolved) return resolved
+      continue
+    }
+    try {
+      if (existsSync(candidate)) return candidate
+    } catch {}
+  }
+  return undefined
+}
+
+function resolveAgentRoot(explicit?: string, hermesHome = detectHermesHome()): string | undefined {
+  const candidates = [
+    explicit,
+    process.env.HERMES_AGENT_ROOT,
+    join(hermesHome, 'hermes-agent'),
+    agentRootFromHermesBin(),
+    process.cwd(),
+    join(process.cwd(), 'hermes-agent'),
+  ].filter((value): value is string => !!value && value.trim().length > 0)
+  return candidates.find(candidate => existsSync(join(candidate, 'run_agent.py')))
+}
+
+function bridgeCommand(options: AgentBridgeManagerOptions): BridgeCommand {
+  const hermesHome = options.hermesHome || detectHermesHome()
+  const agentRoot = resolveAgentRoot(options.agentRoot, hermesHome)
+  const explicitPython = options.python || process.env.HERMES_AGENT_BRIDGE_PYTHON
+  if (explicitPython) {
+    return { command: explicitPython, argsPrefix: [], agentRoot, hermesHome }
+  }
+
+  const venvPython = firstExistingExecutable(pathCandidates(agentRoot))
+  if (venvPython) {
+    return { command: venvPython, argsPrefix: [], agentRoot, hermesHome }
+  }
+
+  const shebangPython = hermesBinPython()
+  if (shebangPython && existsSync(shebangPython)) {
+    return { command: shebangPython, argsPrefix: [], agentRoot, hermesHome }
+  }
+
+  const uv = firstExistingExecutable(uvCandidates(agentRoot))
+  if (uv) {
+    const prefix = ['run']
+    if (agentRoot) prefix.push('--project', agentRoot)
+    prefix.push('python')
+    return { command: uv, argsPrefix: prefix, agentRoot, hermesHome }
+  }
+
+  const fallback = firstExistingExecutable([
+    process.env.PYTHON || '',
+    ...(process.platform === 'win32' ? ['py', 'python', 'python3'] : ['python3', 'python']),
+  ]) || (process.platform === 'win32' ? 'python' : 'python3')
+  return { command: fallback, argsPrefix: [], agentRoot, hermesHome }
+}
+
+function bridgeScriptPath(): string {
+  const candidates = [
+    // Built server: dist/server/index.js -> dist/server/agent-bridge/hermes_bridge.py
+    resolve(__dirname, 'agent-bridge', 'hermes_bridge.py'),
+    // ts-node/dev source tree.
+    resolve(__dirname, 'services/hermes/agent-bridge/hermes_bridge.py'),
+    resolve(process.cwd(), 'packages/server/src/services/hermes/agent-bridge/hermes_bridge.py'),
+  ]
+  const found = candidates.find(candidate => existsSync(candidate))
+  if (!found) {
+    throw new Error(`agent bridge Python script not found. Tried: ${candidates.join(', ')}`)
+  }
+  return found
+}
+
+export class AgentBridgeManager {
+  readonly endpoint: string
+  private readonly options: AgentBridgeManagerOptions
+  private child: ChildProcess | null = null
+  private starting: Promise<void> | null = null
+  private ready = false
+
+  constructor(options: AgentBridgeManagerOptions = {}) {
+    this.options = options
+    this.endpoint = options.endpoint || process.env.HERMES_AGENT_BRIDGE_ENDPOINT || DEFAULT_AGENT_BRIDGE_ENDPOINT
+  }
+
+  get running(): boolean {
+    return !!this.child && !this.child.killed && this.ready
+  }
+
+  async start(): Promise<void> {
+    if (this.running) return
+    if (this.starting) return this.starting
+    this.starting = this.startProcess()
+    try {
+      await this.starting
+    } finally {
+      this.starting = null
+    }
+  }
+
+  private async startProcess(): Promise<void> {
+    const script = bridgeScriptPath()
+    const command = bridgeCommand(this.options)
+    const args = [...command.argsPrefix, script, '--endpoint', this.endpoint]
+    const agentRoot = command.agentRoot
+    const hermesHome = command.hermesHome
+    if (agentRoot) args.push('--agent-root', agentRoot)
+    if (hermesHome) args.push('--hermes-home', hermesHome)
+
+    const env = {
+      ...process.env,
+      HERMES_AGENT_BRIDGE_ENDPOINT: this.endpoint,
+      HERMES_HOME: hermesHome,
+      ...(agentRoot ? { HERMES_AGENT_ROOT: agentRoot } : {}),
+    }
+
+    logger.info('[agent-bridge] starting: %s %s', command.command, args.join(' '))
+    const child = spawn(command.command, args, {
+      env,
+      cwd: process.cwd(),
+      stdio: ['ignore', 'pipe', 'pipe'],
+      windowsHide: true,
+    })
+    this.child = child
+    this.ready = false
+
+    child.once('exit', (code, signal) => {
+      logger.warn('[agent-bridge] exited code=%s signal=%s', code, signal)
+      this.ready = false
+      if (this.child === child) this.child = null
+    })
+
+    child.stderr?.on('data', chunk => {
+      const text = String(chunk).trim()
+      if (text) logger.warn('[agent-bridge] %s', text)
+    })
+
+    await new Promise<void>((resolveReady, rejectReady) => {
+      let buffered = ''
+      const startupTimeoutMs = this.options.startupTimeoutMs
+        ?? envPositiveInt('HERMES_AGENT_BRIDGE_STARTUP_TIMEOUT_MS')
+        ?? DEFAULT_AGENT_BRIDGE_STARTUP_TIMEOUT_MS
+      const timeout = setTimeout(() => {
+        cleanup()
+        rejectReady(new Error(`agent bridge did not become ready within ${startupTimeoutMs}ms`))
+      }, startupTimeoutMs)
+
+      const cleanup = () => {
+        clearTimeout(timeout)
+        child.off('exit', onExitBeforeReady)
+        child.off('error', onError)
+      }
+
+      const onError = (err: Error) => {
+        cleanup()
+        child.stdout?.off('data', onStdout)
+        rejectReady(err)
+      }
+
+      const onExitBeforeReady = (code: number | null, signal: NodeJS.Signals | null) => {
+        cleanup()
+        child.stdout?.off('data', onStdout)
+        rejectReady(new Error(`agent bridge exited before ready code=${code} signal=${signal}`))
+      }
+
+      let readyResolved = false
+      const onStdout = (chunk: Buffer) => {
+        const text = chunk.toString('utf8')
+        buffered += text
+        for (;;) {
+          const newline = buffered.indexOf('\n')
+          if (newline < 0) break
+          const line = buffered.slice(0, newline).trim()
+          buffered = buffered.slice(newline + 1)
+          if (!line) continue
+          logger.info('[agent-bridge] %s', line)
+          if (!readyResolved) {
+            try {
+              const parsed = JSON.parse(line)
+              if (parsed?.event === 'ready') {
+                this.ready = true
+                readyResolved = true
+                cleanup()
+                resolveReady()
+                return
+              }
+            } catch {}
+          }
+        }
+      }
+
+      child.once('error', onError)
+      child.once('exit', onExitBeforeReady)
+      child.stdout?.on('data', onStdout)
+    })
+
+    logger.info('[agent-bridge] ready at %s', this.endpoint)
+  }
+
+  async stop(): Promise<void> {
+    const child = this.child
+    if (!child) return
+    this.ready = false
+    this.child = null
+
+    await new Promise<void>((resolveStop) => {
+      const timeout = setTimeout(() => {
+        if (!child.killed) child.kill('SIGKILL')
+        resolveStop()
+      }, 1500)
+      child.once('exit', () => {
+        clearTimeout(timeout)
+        resolveStop()
+      })
+      if (!child.killed) {
+        child.kill('SIGTERM')
+      }
+    })
+  }
+}
+
+let singleton: AgentBridgeManager | null = null
+
+export function getAgentBridgeManager(): AgentBridgeManager {
+  if (!singleton) singleton = new AgentBridgeManager()
+  return singleton
+}
+
+export async function startAgentBridgeManager(): Promise<AgentBridgeManager> {
+  const manager = getAgentBridgeManager()
+  await manager.start()
+  return manager
+}
@@ -101,6 +101,8 @@ class AgentClient {
            reconnectionAttempts: Infinity,
            reconnectionDelay: 1000,
            reconnectionDelayMax: 30000,
+            randomizationFactor: 0.5,
+            timeout: 30000,
        })

        this.bindEvents()
@@ -424,7 +424,13 @@ export class GroupChatServer {
        const servers = Array.isArray(httpServers) ? httpServers : [httpServers]

        this.io = new Server(servers[0], {
-            cors: { origin: '*' }
+            cors: { origin: '*' },
+            pingInterval: 25_000,
+            pingTimeout: 90_000,
+            connectionStateRecovery: {
+                maxDisconnectionDuration: 2 * 60_000,
+                skipMiddlewares: true,
+            },
        })
        servers.slice(1).forEach((httpServer) => this.io.attach(httpServer))
        this.nsp = this.io.of('/group-chat')
@@ -27,7 +27,7 @@ function shouldStopGatewaysOnShutdown(signal: string): boolean {
  return shouldStop
 }

-export function bindShutdown(server: any, groupChatServer?: any, chatRunServer?: any): void {
+export function bindShutdown(server: any, groupChatServer?: any, chatRunServer?: any, agentBridgeManager?: any): void {
  let isShuttingDown = false

  const shutdown = async (signal: string) => {
@@ -58,6 +58,15 @@ export function bindShutdown(server: any, groupChatServer?: any, chatRunServer?:
        logger.info('Skipping gateway shutdown for %s', signal)
      }

+      if (agentBridgeManager) {
+        try {
+          await agentBridgeManager.stop()
+          logger.info('Agent bridge stopped')
+        } catch (err) {
+          logger.warn(err, 'Failed to stop agent bridge (non-fatal)')
+        }
+      }
+
      // Close ChatRunSocket first to abort all active runs and close EventSource connections
      if (chatRunServer) {
        chatRunServer.close()
@@ -62,13 +62,9 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    base_url: 'https://generativelanguage.googleapis.com/v1beta/openai',
    models: [
      'gemini-3.1-pro-preview',
+      'gemini-3-pro-preview',
      'gemini-3-flash-preview',
      'gemini-3.1-flash-lite-preview',
-      'gemini-2.5-pro',
-      'gemini-2.5-flash',
-      'gemini-2.5-flash-lite',
-      'gemma-4-31b-it',
-      'gemma-4-26b-it',
    ],
  },
  {
@@ -76,7 +72,7 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    value: 'deepseek',
    builtin: true,
    base_url: 'https://api.deepseek.com',
-    models: ['deepseek-v4-flash', 'deepseek-v4-pro'],
+    models: ['deepseek-v4-pro', 'deepseek-v4-flash', 'deepseek-chat', 'deepseek-reasoner'],
  },
  {
    label: 'Z.AI / GLM',
@@ -98,7 +94,6 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    builtin: true,
    base_url: 'https://api.kimi.com/coding/v1',
    models: [
-      'kimi-for-coding',
      'kimi-k2.6',
      'kimi-k2.5',
      'kimi-k2-thinking',
@@ -124,7 +119,17 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    value: 'xai',
    builtin: true,
    base_url: 'https://api.x.ai/v1',
-    models: ['grok-4.20-reasoning', 'grok-4-1-fast-reasoning'],
+    models: [
+      'grok-4.20-0309-reasoning',
+      'grok-4.20-0309-non-reasoning',
+      'grok-4.20-multi-agent-0309',
+      'grok-4-1-fast',
+      'grok-4-1-fast-non-reasoning',
+      'grok-4-fast',
+      'grok-4-fast-non-reasoning',
+      'grok-4',
+      'grok-code-fast-1',
+    ],
  },
  {
    label: 'MiniMax',
@@ -146,12 +151,13 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    builtin: true,
    base_url: 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1',
    models: [
+      'qwen3.6-plus',
+      'kimi-k2.5',
      'qwen3.5-plus',
      'qwen3-coder-plus',
      'qwen3-coder-next',
      'glm-5',
      'glm-4.7',
-      'kimi-k2.5',
      'MiniMax-M2.5',
    ],
  },
@@ -166,13 +172,13 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    // returns HTTP 401 for those keys.
    base_url: 'https://coding-intl.dashscope.aliyuncs.com/v1',
    models: [
+      'qwen3.6-plus',
      'qwen3.5-plus',
-      'qwen3-max-2026-01-23',
-      'qwen3-coder-next',
      'qwen3-coder-plus',
+      'qwen3-coder-next',
+      'kimi-k2.5',
      'glm-5',
      'glm-4.7',
-      'kimi-k2.5',
      'MiniMax-M2.5',
    ],
  },
@@ -182,14 +188,15 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    builtin: true,
    base_url: 'https://router.huggingface.co/v1',
    models: [
+      'moonshotai/Kimi-K2.5',
      'Qwen/Qwen3.5-397B-A17B',
      'Qwen/Qwen3.5-35B-A3B',
      'deepseek-ai/DeepSeek-V3.2',
-      'moonshotai/Kimi-K2.5',
      'MiniMaxAI/MiniMax-M2.5',
      'zai-org/GLM-5',
      'XiaomiMiMo/MiMo-V2-Flash',
      'moonshotai/Kimi-K2-Thinking',
+      'moonshotai/Kimi-K2.6',
    ],
  },
  {
@@ -214,14 +221,11 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    builtin: true,
    base_url: 'https://api.xiaomimimo.com/v1',
    models: [
-      'mimo-v2-omni',
-      'mimo-v2-pro',
-      'mimo-v2-tts',
-      'mimo-v2.5',
      'mimo-v2.5-pro',
-      'mimo-v2.5-tts',
-      'mimo-v2.5-tts-voiceclone',
-      'mimo-v2.5-tts-voicedesign',
+      'mimo-v2.5',
+      'mimo-v2-pro',
+      'mimo-v2-omni',
+      'mimo-v2-flash',
    ],
  },
  {
@@ -243,18 +247,21 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    builtin: true,
    base_url: 'https://ai-gateway.vercel.sh/v1',
    models: [
-      'anthropic/claude-opus-4.6',
+      'moonshotai/kimi-k2.6',
+      'alibaba/qwen3.6-plus',
+      'zai/glm-5.1',
+      'minimax/minimax-m2.7',
      'anthropic/claude-sonnet-4.6',
-      'anthropic/claude-sonnet-4.5',
+      'anthropic/claude-opus-4.7',
+      'anthropic/claude-opus-4.6',
      'anthropic/claude-haiku-4.5',
-      'openai/gpt-5',
-      'openai/gpt-4.1',
-      'openai/gpt-4.1-mini',
-      'google/gemini-3-pro-preview',
+      'openai/gpt-5.4',
+      'openai/gpt-5.4-mini',
+      'openai/gpt-5.3-codex',
+      'google/gemini-3.1-pro-preview',
      'google/gemini-3-flash',
-      'google/gemini-2.5-pro',
-      'google/gemini-2.5-flash',
-      'deepseek/deepseek-v3.2',
+      'google/gemini-3.1-flash-lite-preview',
+      'xai/grok-4.20-reasoning',
    ],
  },
  {
@@ -277,10 +284,10 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    builtin: true,
    base_url: 'https://opencode.ai/zen/v1',
    models: [
+      'kimi-k2.5',
      'gpt-5.4-pro',
      'gpt-5.4',
      'gpt-5.3-codex',
-      'gpt-5.3-codex-spark',
      'gpt-5.2',
      'gpt-5.2-codex',
      'gpt-5.1',
@@ -308,7 +315,6 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
      'glm-5',
      'glm-4.7',
      'glm-4.6',
-      'kimi-k2.5',
      'kimi-k2-thinking',
      'kimi-k2',
      'qwen3-coder',
@@ -320,7 +326,20 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
    value: 'opencode-go',
    builtin: true,
    base_url: 'https://opencode.ai/zen/go/v1',
-    models: ['glm-5.1', 'glm-5', 'kimi-k2.5', 'mimo-v2-pro', 'mimo-v2-omni', 'minimax-m2.7', 'minimax-m2.5'],
+    models: [
+      'kimi-k2.6',
+      'kimi-k2.5',
+      'glm-5.1',
+      'glm-5',
+      'mimo-v2.5-pro',
+      'mimo-v2.5',
+      'mimo-v2-pro',
+      'mimo-v2-omni',
+      'minimax-m2.7',
+      'minimax-m2.5',
+      'qwen3.6-plus',
+      'qwen3.5-plus',
+    ],
  },
  {
    label: 'LongCat',
@@ -352,12 +371,13 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
      'moonshotai/kimi-k2.6',
      'xiaomi/mimo-v2.5-pro',
      'xiaomi/mimo-v2.5',
+      'tencent/hy3-preview',
      'anthropic/claude-opus-4.7',
      'anthropic/claude-opus-4.6',
      'anthropic/claude-sonnet-4.6',
      'anthropic/claude-sonnet-4.5',
      'anthropic/claude-haiku-4.5',
-      'openai/gpt-5.4',
+      'openai/gpt-5.5',
      'openai/gpt-5.4-mini',
      'openai/gpt-5.3-codex',
      'google/gemini-3-pro-preview',
@@ -374,10 +394,12 @@ export const PROVIDER_PRESETS: ProviderPreset[] = [
      'z-ai/glm-5v-turbo',
      'z-ai/glm-5-turbo',
      'x-ai/grok-4.20-beta',
+      'x-ai/grok-4.3',
      'nvidia/nemotron-3-super-120b-a12b',
      'arcee-ai/trinity-large-thinking',
-      'openai/gpt-5.4-pro',
+      'openai/gpt-5.5-pro',
      'openai/gpt-5.4-nano',
+      'deepseek/deepseek-v4-pro',
    ],
  },
  {
@@ -1,7 +1,7 @@
 import * as esbuild from 'esbuild'
 import { resolve, dirname } from 'path'
 import { fileURLToPath } from 'url'
-import { readFileSync } from 'fs'
+import { cpSync, mkdirSync, readFileSync } from 'fs'

 const rootDir = resolve(dirname(fileURLToPath(import.meta.url)), '..')
 const pkg = JSON.parse(readFileSync(resolve(rootDir, 'package.json'), 'utf-8'))
@@ -23,3 +23,10 @@ await esbuild.build({
  treeShaking: true,
  logLevel: 'info',
 })
+
+const bridgeOutDir = resolve(rootDir, 'dist/server/agent-bridge')
+mkdirSync(bridgeOutDir, { recursive: true })
+cpSync(
+  resolve(rootDir, 'packages/server/src/services/hermes/agent-bridge/hermes_bridge.py'),
+  resolve(bridgeOutDir, 'hermes_bridge.py'),
+)
@@ -4,10 +4,6 @@ import {
  PROVIDER_PRESETS as SERVER_PROVIDER_PRESETS,
  buildProviderModelMap as buildServerProviderModelMap,
 } from '../../packages/server/src/shared/providers'
-import {
-  PROVIDER_PRESETS as CLIENT_PROVIDER_PRESETS,
-  buildProviderModelMap as buildClientProviderModelMap,
-} from '../../packages/client/src/shared/providers'

 const OPENAI_CODEX_PROVIDER = 'openai-codex'
 const GPT_5_5_MODEL = 'gpt-5.5'
@@ -19,13 +15,11 @@ function modelsForProvider(providerPresets: Array<{ value: string; models: strin
 }

 describe('provider presets', () => {
-  it('lists GPT-5.5 for OpenAI Codex on both client and server', () => {
-    expect(modelsForProvider(CLIENT_PROVIDER_PRESETS, OPENAI_CODEX_PROVIDER)).toContain(GPT_5_5_MODEL)
+  it('lists GPT-5.5 for OpenAI Codex', () => {
    expect(modelsForProvider(SERVER_PROVIDER_PRESETS, OPENAI_CODEX_PROVIDER)).toContain(GPT_5_5_MODEL)
  })

  it('exposes GPT-5.5 through provider model maps', () => {
-    expect(buildClientProviderModelMap()[OPENAI_CODEX_PROVIDER]).toContain(GPT_5_5_MODEL)
    expect(buildServerProviderModelMap()[OPENAI_CODEX_PROVIDER]).toContain(GPT_5_5_MODEL)
  })
 })