Merge branch 'feature/browser-use-api' into develop

# Conflicts:
#	.env.example
#	browser_env/entrypoint.sh
#	docker-compose.yml
#	hermes_code/tools/browser_use_tool.py

.env.example

@@ -13,4 +13,10 @@ TELEGRAM_ALLOWED_USERS=
 TELEGRAM_HOME_CHANNEL=
 
 BROWSER_URL=http://browser:9222
-BROWSER_VIEW_URL=http://localhost:6080
+BROWSER_VIEW_URL=
+
+BROWSER_API_HOST=0.0.0.0
+BROWSER_API_PORT=8088
+BROWSER_USE_RPC_URL=http://browser:8787/run
+BROWSER_USE_RPC_TIMEOUT=900
+BROWSER_API_MAX_CONCURRENCY=2

api/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
+
+WORKDIR /app
+
+COPY requirements.txt /app/requirements.txt
+RUN pip install --no-cache-dir -r /app/requirements.txt
+
+COPY . /app/api
+
+CMD ["uvicorn", "api.main:app", "--host", "0.0.0.0", "--port", "8088"]

api/README.md

@@ -0,0 +1,120 @@
+# Browser REST API
+
+REST API-обертка над `browser-use` RPC (`POST /run` в контейнере браузера).
+
+Сервис принимает задачу, ставит ее в in-memory очередь, выполняет через `browser-use` и отдает статус/результат по `task_id`.
+
+## Актуальный статус
+
+Проверено smoke-тестом:
+- `GET /health` отвечает `200` с `{"ok": true}`
+- `POST /api/browser/tasks` возвращает `202` и `task_id`
+- `GET /api/browser/tasks/{task_id}` возвращает `queued/running/...`
+- `GET /api/browser/tasks/{task_id}/result` возвращает `202`, пока задача не завершена
+
+## Архитектура
+
+Слои сейчас разделены и выглядят нормально для MVP:
+
+- `api/main.py` — точка входа ASGI (`uvicorn api.main:app`), сборка `FastAPI` и lifespan
+- `api/routes/tasks.py` — HTTP-слой (валидация входа/выхода, status codes)
+- `api/services/task_service.py` — orchestration (фоновые задачи, timeout, обработка ошибок)
+- `api/repositories/task_store.py` — in-memory хранилище задач
+- `api/clients/browser_rpc_client.py` — aiohttp-клиент к browser RPC
+- `api/clients/browser_rpc_contracts.py` — protocol + исключения RPC-слоя
+- `api/contracts/task_schemas.py` — Pydantic request/response DTO
+- `api/domain/task_status.py` — доменный enum статусов
+- `api/core/settings.py` — конфигурация из env
+
+## Ограничения текущей реализации
+
+- хранилище in-memory: после рестарта контейнера задачи теряются
+- нет ретраев RPC при транспортных ошибках
+- нет отмены задач через API
+- один инстанс процесса хранит задачи только локально (без shared state)
+
+## Переменные окружения
+
+- `BROWSER_API_HOST` (default: `0.0.0.0`)
+- `BROWSER_API_PORT` (default: `8080`)
+- `BROWSER_USE_RPC_URL` (default: `http://browser:8787/run`)
+- `BROWSER_USE_RPC_TIMEOUT` (default: `900`)
+- `BROWSER_API_MAX_CONCURRENCY` (default: `2`)
+
+## Локальный запуск
+
+```zsh
+cd "/Users/fedorkobylkevic/PycharmProjects/BrowserUse_and_ComputerUse_skills"
+source .venv/bin/activate
+uvicorn api.main:app --host 0.0.0.0 --port 8088
+```
+
+## Запуск через Docker Compose
+
+```zsh
+cd "/Users/fedorkobylkevic/PycharmProjects/BrowserUse_and_ComputerUse_skills"
+docker compose build browser-api
+docker compose up -d browser browser-api
+docker compose logs -f browser-api
+```
+
+## REST API
+
+### `GET /health`
+
+Проверка доступности API.
+
+Пример ответа:
+
+```json
+{"ok": true}
+```
+
+### `POST /api/browser/tasks`
+
+Создать задачу.
+
+Request:
+
+```json
+{
+  "task": "Открой example.com и верни title",
+  "timeout": 300,
+  "metadata": {"source": "manual"}
+}
+```
+
+Response `202`:
+
+```json
+{
+  "task_id": "53f54fa4c1f24219b3949d56b0457875",
+  "status": "queued"
+}
+```
+
+### `GET /api/browser/tasks/{task_id}`
+
+Текущий статус и таймстемпы.
+
+### `GET /api/browser/tasks/{task_id}/result`
+
+- `202` если задача еще `queued/running`
+- `200` с финальным payload после завершения
+
+## Быстрый end-to-end пример
+
+```zsh
+curl -sS http://localhost:8088/health
+
+RESP=$(curl -sS -X POST http://localhost:8088/api/browser/tasks \
+  -H "Content-Type: application/json" \
+  -d '{"task":"Открой example.com и верни title","timeout":30}')
+
+echo "$RESP"
+
+TASK_ID=$(python -c "import json,sys;print(json.loads(sys.argv[1])['task_id'])" "$RESP")
+
+curl -sS "http://localhost:8088/api/browser/tasks/$TASK_ID"
+curl -sS "http://localhost:8088/api/browser/tasks/$TASK_ID/result"
+```

api/clients/browser_rpc_client.py

@@ -0,0 +1,38 @@
+from typing import Any
+
+import aiohttp
+
+from api.clients.browser_rpc_contracts import BrowserRpcError
+
+
+class BrowserRpcClient:
+    def __init__(self, rpc_url: str, session: aiohttp.ClientSession) -> None:
+        self._rpc_url = rpc_url
+        self._session = session
+
+    async def run(self, task: str, timeout_sec: float) -> dict[str, Any]:
+        payload = {"task": task}
+        timeout = aiohttp.ClientTimeout(total=timeout_sec)
+
+        try:
+            async with self._session.post(self._rpc_url, json=payload, timeout=timeout) as response:
+                if response.status >= 400:
+                    body = await response.text()
+                    raise BrowserRpcError(f"RPC HTTP: {response.status}: {body}")
+
+                try:
+                    data = await response.json(content_type=None)
+                except aiohttp.ContentTypeError as exc:
+                    raise BrowserRpcError("RPC returned non-JSON response") from exc
+        except aiohttp.ClientError as exc:
+            raise BrowserRpcError(f"Transport error: {exc}") from exc
+
+        if not isinstance(data, dict):
+            raise BrowserRpcError("RPC returned invalid payload type")
+
+        return data
+
+
+async def run_browser_task(rpc_url: str, task: str, timeout_sec: float) -> dict[str, Any]:
+    async with aiohttp.ClientSession() as session:
+        return await BrowserRpcClient(rpc_url, session=session).run(task=task, timeout_sec=timeout_sec)

api/clients/browser_rpc_contracts.py

@@ -0,0 +1,8 @@
+from typing import Any, Protocol
+
+
+class BrowserRpcError(RuntimeError): ...
+
+
+class BrowserRpcRunner(Protocol):
+    async def run(self, task: str, timeout_sec: float) -> dict[str, Any]: ...

api/contracts/task_schemas.py

@@ -0,0 +1,43 @@
+from typing import Any
+
+from pydantic import BaseModel, Field
+
+from api.domain.task_status import TaskStatus
+
+
+class BrowserTaskRequest(BaseModel):
+    """Запрос на запуск задачи в browser-use агенте."""
+
+    task: str = Field(..., description="Текстовая задача для browser-use агента")
+    timeout: int = Field(300, description="Максимальное время выполнения задачи в секундах")
+    metadata: dict[str, Any] | None = Field(default=None, description="Дополнительные метаданные клиента")
+
+
+class BrowserTaskAcceptedResponse(BaseModel):
+    """Ответ о том, что задача принята в обработку."""
+
+    task_id: str
+    status: TaskStatus
+
+
+class BrowserTaskStatusResponse(BaseModel):
+    """Текущий статус задачи и временные отметки ее выполнения."""
+
+    task_id: str
+    status: TaskStatus
+    create_at: float = Field(..., description="Время создания задачи в Unix timestamp")
+    started_at: float | None = Field(default=None, description="Время начала выполнения в Unix timestamp")
+    finished_at: float | None = Field(default=None, description="Время завершения выполнения в Unix timestamp")
+    error: str | None = Field(default=None, description="Текст ошибки, если задача завершилась с ошибкой")
+
+
+class BrowserTaskResultResponse(BaseModel):
+    """Финальный результат выполнения задачи в browser-use."""
+
+    task_id: str
+    status: TaskStatus
+    success: bool = Field(..., description="Успешно ли выполнена задача")
+    execution_time: float = Field(..., description="Фактическое время выполнения в секундах")
+    result: str | None = Field(default=None, description="Итоговый текстовый результат")
+    error: str | None = Field(default=None, description="Текст ошибки, если выполнение не удалось")
+    raw_response: dict[str, Any] | None = Field(default=None, description="Сырой ответ от browser-use RPC")

api/core/settings.py

@@ -0,0 +1,16 @@
+import os
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class Settings:
+    app_host: str = os.getenv("BROWSER_API_HOST", "0.0.0.0")
+    app_port: int = int(os.getenv("BROWSER_API_PORT", "8080"))
+
+    browser_rpc_url: str = os.getenv("BROWSER_USE_RPC_URL", "http://browser:8787/run")
+    browser_rpc_timeout: float = float(os.getenv("BROWSER_USE_RPC_TIMEOUT", "900"))
+
+    max_concurrency: int = int(os.getenv("BROWSER_API_MAX_CONCURRENCY", "2"))
+
+
+settings = Settings()

api/domain/task_status.py

@@ -0,0 +1,9 @@
+from enum import Enum
+
+
+class TaskStatus(str, Enum):
+    """Состояние задачи браузерного агента."""
+    queued = "queued"
+    running = "running"
+    succeeded = "succeeded"
+    failed = "failed"

api/main.py

@@ -0,0 +1,46 @@
+from contextlib import asynccontextmanager
+
+import aiohttp
+from fastapi import FastAPI
+
+from api.clients.browser_rpc_client import BrowserRpcClient
+from api.core.settings import settings
+from api.repositories.task_store import TaskStore
+from api.routes.tasks import router as tasks_router
+from api.services.task_service import TaskService
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    session = aiohttp.ClientSession()
+    task_service = TaskService(
+        store=TaskStore(),
+        rpc_client=BrowserRpcClient(settings.browser_rpc_url, session=session),
+        max_concurrency=settings.max_concurrency,
+        rpc_timeout_cap=settings.browser_rpc_timeout,
+    )
+    app.state.task_service = task_service
+    try:
+        yield
+    finally:
+        await task_service.close()
+        await session.close()
+
+
+def create_app() -> FastAPI:
+    app = FastAPI(
+        title="Browser API",
+        version="1.0.0",
+        description="REST API for submitting tasks to browser-use and retrieving their status/results.",
+        lifespan=lifespan,
+    )
+    app.include_router(tasks_router)
+
+    @app.get("/health")
+    async def health() -> dict:
+        return {"ok": True}
+
+    return app
+
+
+app = create_app()

api/repositories/task_store.py

@@ -0,0 +1,76 @@
+import time
+import uuid
+from asyncio import Lock
+from dataclasses import dataclass, field
+from typing import Any
+
+from api.domain.task_status import TaskStatus
+
+
+@dataclass
+class TaskRecord:
+    task_id: str
+    task: str
+    timeout: int
+    metadata: dict[str, Any] | None
+    status: TaskStatus = TaskStatus.queued
+    create_at: float = field(default_factory=time.time)
+    started_at: float | None = None
+    finished_at: float | None = None
+    result: str | None = None
+    error: str | None = None
+    raw_response: dict[str, Any] | None = None
+
+    @property
+    def execution_time(self) -> float:
+        if self.started_at is None:
+            return 0
+        end = self.finished_at if self.finished_at is not None else time.time()
+        return max(0, end - self.started_at)
+
+
+class TaskStore:
+    def __init__(self) -> None:
+        self._lock = Lock()
+        self._tasks: dict[str, TaskRecord] = {}
+
+    async def create(self, task: str, timeout: int, metadata: dict[str, Any] | None) -> TaskRecord:
+        task_id = uuid.uuid4().hex
+        rec = TaskRecord(task_id=task_id, task=task, timeout=timeout, metadata=metadata)
+        async with self._lock:
+            self._tasks[task_id] = rec
+        return rec
+
+    async def get(self, task_id: str) -> TaskRecord | None:
+        async with self._lock:
+            return self._tasks.get(task_id)
+
+    async def set_running(self, task_id: str) -> TaskRecord | None:
+        async with self._lock:
+            rec = self._tasks.get(task_id)
+            if rec is None:
+                return None
+            rec.status = TaskStatus.running
+            rec.started_at = time.time()
+            return rec
+
+    async def set_done(
+            self,
+            task_id: str,
+            success: bool,
+            raw_response: dict[str, Any] | None,
+            error: str | None,
+            result: str | None = None,
+    ) -> TaskRecord | None:
+        async with self._lock:
+            rec = self._tasks.get(task_id)
+            if rec is None:
+                return None
+            rec.finished_at = time.time()
+            rec.raw_response = raw_response
+            rec.error = error if error is not None else (
+                raw_response.get("error") if isinstance(raw_response, dict) else None)
+            rec.result = result if result is not None else (
+                raw_response.get("result") if isinstance(raw_response, dict) else None)
+            rec.status = TaskStatus.succeeded if success else TaskStatus.failed
+            return rec

api/requirements.txt

@@ -0,0 +1,4 @@
+fastapi==0.135.3
+uvicorn[standard]==0.44.0
+aiohttp==3.13.5
+pydantic==2.12.5

api/routes/tasks.py

@@ -0,0 +1,80 @@
+from fastapi import APIRouter, Depends, HTTPException, Request
+from fastapi.responses import JSONResponse
+
+from api.contracts.task_schemas import (
+    BrowserTaskAcceptedResponse,
+    BrowserTaskRequest,
+    BrowserTaskResultResponse,
+    BrowserTaskStatusResponse,
+)
+from api.domain.task_status import TaskStatus
+from api.repositories.task_store import TaskRecord
+from api.services.task_service import TaskService
+
+router = APIRouter(prefix="/api/browser", tags=["browser-tasks"])
+
+
+def get_task_service(request: Request) -> TaskService:
+    return request.app.state.task_service
+
+
+@router.post("/tasks", response_model=BrowserTaskAcceptedResponse, status_code=202)
+async def create_task(
+        payload: BrowserTaskRequest,
+        service: TaskService = Depends(get_task_service),
+) -> BrowserTaskAcceptedResponse:
+    rec = await service.submit_task(task=payload.task.strip(), timeout=payload.timeout, metadata=payload.metadata)
+    return BrowserTaskAcceptedResponse(task_id=rec.task_id, status=rec.status)
+
+
+@router.get("/tasks/{task_id}", response_model=BrowserTaskStatusResponse)
+async def get_task_status(task_id: str, service: TaskService = Depends(get_task_service)) -> BrowserTaskStatusResponse:
+    rec = await service.get_task(task_id)
+    if rec is None:
+        raise HTTPException(status_code=404, detail="Task not found")
+    return _to_status_response(rec)
+
+
+@router.get("/tasks/{task_id}/result", response_model=BrowserTaskResultResponse)
+async def get_task_result(
+        task_id: str,
+        service: TaskService = Depends(get_task_service),
+) -> JSONResponse | BrowserTaskResultResponse:
+    rec = await service.get_task(task_id)
+    if rec is None:
+        raise HTTPException(status_code=404, detail="Task not found")
+
+    if rec.status in (TaskStatus.queued, TaskStatus.running):
+        return JSONResponse(
+            status_code=202,
+            content={
+                "task_id": rec.task_id,
+                "status": rec.status.value,
+                "success": False,
+                "execution_time": rec.execution_time,
+                "result": None,
+                "error": None,
+                "raw_response": None,
+            },
+        )
+
+    return BrowserTaskResultResponse(
+        task_id=rec.task_id,
+        status=rec.status,
+        success=(rec.status == TaskStatus.succeeded),
+        execution_time=rec.execution_time,
+        result=rec.result,
+        error=rec.error,
+        raw_response=rec.raw_response,
+    )
+
+
+def _to_status_response(rec: TaskRecord) -> BrowserTaskStatusResponse:
+    return BrowserTaskStatusResponse(
+        task_id=rec.task_id,
+        status=rec.status,
+        create_at=rec.create_at,
+        started_at=rec.started_at,
+        finished_at=rec.finished_at,
+        error=rec.error,
+    )

api/services/task_service.py

@@ -0,0 +1,83 @@
+import asyncio
+
+from api.clients.browser_rpc_contracts import BrowserRpcError, BrowserRpcRunner
+from api.repositories.task_store import TaskRecord, TaskStore
+
+
+class TaskService:
+    def __init__(
+            self,
+            store: TaskStore,
+            rpc_client: BrowserRpcRunner,
+            max_concurrency: int,
+            rpc_timeout_cap: float | None = None,
+    ) -> None:
+        self._store = store
+        self._rpc_client = rpc_client
+        self._semaphore = asyncio.Semaphore(max_concurrency)
+        self._rpc_timeout_cap = rpc_timeout_cap
+        self._background_tasks: set[asyncio.Task[None]] = set()
+
+    async def submit_task(self, task: str, timeout: int, metadata: dict | None) -> TaskRecord:
+        record = await self._store.create(task=task, timeout=timeout, metadata=metadata)
+        background_task = asyncio.create_task(self._worker(record.task_id))
+        self._background_tasks.add(background_task)
+        background_task.add_done_callback(self._background_tasks.discard)
+        return record
+
+    async def get_task(self, task_id: str) -> TaskRecord | None:
+        return await self._store.get(task_id)
+
+    async def close(self) -> None:
+        if not self._background_tasks:
+            return
+
+        for task in list(self._background_tasks):
+            task.cancel()
+        await asyncio.gather(*self._background_tasks, return_exceptions=True)
+        self._background_tasks.clear()
+
+    async def _worker(self, task_id: str) -> None:
+        rec = await self._store.set_running(task_id)
+        if rec is None:
+            return
+
+        async with self._semaphore:
+            try:
+                rpc_timeout = float(rec.timeout)
+                if self._rpc_timeout_cap is not None:
+                    rpc_timeout = min(rpc_timeout, self._rpc_timeout_cap)
+
+                raw = await asyncio.wait_for(
+                    self._rpc_client.run(task=rec.task, timeout_sec=rpc_timeout),
+                    timeout=float(rec.timeout) + 5,
+                )
+                success = bool(raw.get("success"))
+                await self._store.set_done(
+                    task_id=task_id,
+                    success=success,
+                    raw_response=raw,
+                    error=None,
+                    result=raw.get("result") if isinstance(raw, dict) else None,
+                )
+            except asyncio.TimeoutError:
+                await self._store.set_done(
+                    task_id=task_id,
+                    success=False,
+                    raw_response=None,
+                    error="Timeout exceeded",
+                )
+            except BrowserRpcError as exc:
+                await self._store.set_done(
+                    task_id=task_id,
+                    success=False,
+                    raw_response=None,
+                    error=str(exc),
+                )
+            except Exception as exc:
+                await self._store.set_done(
+                    task_id=task_id,
+                    success=False,
+                    raw_response=None,
+                    error=f"Internal error: {exc}",
+                )

browser_env/Dockerfile.browser

@@ -4,6 +4,8 @@ ENV DEBIAN_FRONTEND=noninteractive
 
 RUN apt-get update && apt-get install -y --no-install-recommends \
     chromium \
+    python3 \
+    python3-pip \
     xvfb \
     fluxbox \
     x11vnc \
@@ -19,9 +21,14 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 WORKDIR /src
 RUN mkdir -p /src/browser_data
 
+RUN python3 -m pip install --no-cache-dir --break-system-packages \
+    "browser-use>=0.12.5" \
+    "langchain-openai>=0.3.0"
+
 COPY entrypoint.sh /entrypoint.sh
+COPY browser_use_runner.py /src/browser_use_runner.py
 RUN chmod +x /entrypoint.sh
 
-EXPOSE 6080 9222
+EXPOSE 6080 9222 8787
 
 ENTRYPOINT ["/entrypoint.sh"]

browser_env/browser_use_runner.py

@@ -0,0 +1,101 @@
+import asyncio
+import json
+import os
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from urllib import error, request
+
+from browser_use import Agent, Browser, ChatOpenAI
+
+
+def _json_response(handler, status_code, payload):
+    data = json.dumps(payload, ensure_ascii=False).encode("utf-8")
+    handler.send_response(status_code)
+    handler.send_header("Content-Type", "application/json; charset=utf-8")
+    handler.send_header("Content-Length", str(len(data)))
+    handler.end_headers()
+    handler.wfile.write(data)
+
+
+async def run_browser_task(task):
+    cdp_url = os.getenv("BROWSER_CDP_URL", "http://127.0.0.1:9222")
+    browser_view_url = os.getenv("BROWSER_VIEW_URL", "")
+
+    browser = Browser(cdp_url=cdp_url)
+
+    llm = ChatOpenAI(
+        model=os.getenv("MODEL_DEFAULT", "qwen3.5-122b"),
+        api_key=os.getenv("OPENAI_API_KEY"),
+        base_url=os.getenv("OPENAI_BASE_URL"),
+        temperature=0.0,
+    )
+
+    agent = Agent(task=task, llm=llm, browser=browser)
+
+    try:
+        history = await agent.run()
+        return {
+            "success": True,
+            "result": history.final_result(),
+            "browser_view": browser_view_url,
+        }
+    except Exception as err:
+        return {"success": False, "error": f"Browser automation failed: {err}"}
+    finally:
+        try:
+            await browser.close()
+        except Exception:
+            pass
+
+
+class BrowserUseRPCHandler(BaseHTTPRequestHandler):
+    def do_GET(self):
+        if self.path != "/health":
+            _json_response(self, 404, {"success": False, "error": "Not found"})
+            return
+
+        try:
+            debug_url = os.getenv("BROWSER_HEALTH_URL", "http://127.0.0.1:9222/json/version")
+            with request.urlopen(debug_url, timeout=2):
+                pass
+            _json_response(self, 200, {"success": True})
+        except Exception as err:
+            _json_response(self, 503, {"success": False, "error": f"Browser is not ready: {err}"})
+
+    def do_POST(self):
+        if self.path != "/run":
+            _json_response(self, 404, {"success": False, "error": "Not found"})
+            return
+
+        try:
+            content_length = int(self.headers.get("Content-Length", "0"))
+            raw = self.rfile.read(content_length)
+            payload = json.loads(raw.decode("utf-8") if raw else "{}")
+            task = payload.get("task", "")
+            if not isinstance(task, str) or not task.strip():
+                _json_response(self, 400, {"success": False, "error": "Field 'task' is required"})
+                return
+
+            result = asyncio.run(run_browser_task(task.strip()))
+            code = 200 if result.get("success") else 500
+            _json_response(self, code, result)
+        except json.JSONDecodeError:
+            _json_response(self, 400, {"success": False, "error": "Invalid JSON payload"})
+        except error.URLError as err:
+            _json_response(self, 503, {"success": False, "error": f"Transport error: {err}"})
+        except Exception as err:
+            _json_response(self, 500, {"success": False, "error": f"Internal error: {err}"})
+
+    def log_message(self, format_str, *args):
+        return
+
+
+def main():
+    host = os.getenv("BROWSER_USE_RPC_HOST", "0.0.0.0")
+    port = int(os.getenv("BROWSER_USE_RPC_PORT", "8787"))
+    server = ThreadingHTTPServer((host, port), BrowserUseRPCHandler)
+    print(f"browser-use RPC listening on {host}:{port}")
+    server.serve_forever()
+
+
+if __name__ == "__main__":
+    main()

browser_env/entrypoint.sh

@@ -1,41 +1,197 @@
-#!/bin/bash
+#!/usr/bin/env bash
+set -Eeuo pipefail
 
-export DISPLAY=:99
+export DISPLAY="${DISPLAY:-:99}"
+DISPLAY_NUM="${DISPLAY#:}"
+XVFB_LOG="/tmp/xvfb.log"
 
-mkdir -p /var/run/dbus
+VNC_PORT="${VNC_PORT:-5900}"
-dbus-uuidgen > /var/lib/dbus/machine-id
+NOVNC_PORT="${NOVNC_PORT:-6080}"
-dbus-daemon --config-file=/usr/share/dbus-1/system.conf --print-address &
+CHROME_LOCAL_DEBUG_PORT="${CHROME_LOCAL_DEBUG_PORT:-9223}"
+CHROME_PUBLIC_DEBUG_PORT="${CHROME_PUBLIC_DEBUG_PORT:-9222}"
+BROWSER_USE_RPC_PORT="${BROWSER_USE_RPC_PORT:-8787}"
+CHROME_PROFILE_DIR="${CHROME_PROFILE_DIR:-/src/browser_data}"
 
-Xvfb :99 -screen 0 1280x720x16 -ac +extension GLX +render -noreset &
+MAX_RESTARTS="${MAX_RESTARTS:-10}"
-sleep 2
+RESTART_WINDOW_SEC="${RESTART_WINDOW_SEC:-60}"
+RESTART_BACKOFF_SEC="${RESTART_BACKOFF_SEC:-2}"
 
-fluxbox &
+PIDS=()
-x11vnc -display :99 -nopw -listen 0.0.0.0 -xkb -forever -shared &
+STOPPING=0
-websockify --web=/usr/share/novnc/ 6080 localhost:5900 &
+WINDOW_START="$(date +%s)"
+RESTART_COUNT=0
 
-socat TCP-LISTEN:9222,fork,reuseaddr TCP:127.0.0.1:9223 &
+log() {
+  printf '[%s] %s\n' "$(date '+%Y-%m-%d %H:%M:%S')" "$*"
+}
 
-echo "--- Запуск Chromium в режиме Local-Only (Port 9223) ---"
+start_bg() {
+  "$@" &
+  local pid=$!
+  PIDS+=("$pid")
+  log "started: $* (pid=$pid)"
+}
+
+wait_for_port() {
+  local host=$1
+  local port=$2
+  local timeout_sec=$3
+  local end_ts=$(( $(date +%s) + timeout_sec ))
+
+  while [ "$(date +%s)" -lt "$end_ts" ]; do
+    if bash -c "</dev/tcp/${host}/${port}" >/dev/null 2>&1; then
+      return 0
+    fi
+    sleep 0.2
+  done
+  return 1
+}
+
+wait_for_x_display() {
+  local timeout_sec=$1
+  local end_ts=$(( $(date +%s) + timeout_sec ))
+
+  while [ "$(date +%s)" -lt "$end_ts" ]; do
+    if [ -S "/tmp/.X11-unix/X${DISPLAY_NUM}" ] && DISPLAY="$DISPLAY" bash -c 'echo >/dev/null' >/dev/null 2>&1; then
+      return 0
+    fi
+    sleep 0.2
+  done
+  return 1
+}
+
+cleanup() {
+  if [ "$STOPPING" -eq 1 ]; then
+    return
+  fi
+  STOPPING=1
+
+  log "shutdown signal received, stopping processes..."
+
+  if [ -n "${CHROME_PID:-}" ] && kill -0 "$CHROME_PID" >/dev/null 2>&1; then
+    kill "$CHROME_PID" >/dev/null 2>&1 || true
+  fi
+
+  for pid in "${PIDS[@]:-}"; do
+    kill "$pid" >/dev/null 2>&1 || true
+  done
+
+  sleep 1
+
+  if [ -n "${CHROME_PID:-}" ] && kill -0 "$CHROME_PID" >/dev/null 2>&1; then
+    kill -9 "$CHROME_PID" >/dev/null 2>&1 || true
+  fi
+
+  for pid in "${PIDS[@]:-}"; do
+    if kill -0 "$pid" >/dev/null 2>&1; then
+      kill -9 "$pid" >/dev/null 2>&1 || true
+    fi
+  done
+
+  log "shutdown complete"
+}
+
+trap cleanup SIGTERM SIGINT EXIT
+
+mkdir -p /var/run/dbus /var/lib/dbus "$CHROME_PROFILE_DIR"
+if [ ! -f /var/lib/dbus/machine-id ]; then
+  dbus-uuidgen > /var/lib/dbus/machine-id 2>/dev/null || true
+fi
+
+# Удаляем stale lock/socket от прошлых падений Xvfb на том же DISPLAY.
+rm -f "/tmp/.X${DISPLAY_NUM}-lock" "/tmp/.X11-unix/X${DISPLAY_NUM}" || true
+
+log "starting X stack on DISPLAY=${DISPLAY}"
+Xvfb "$DISPLAY" -screen 0 1280x720x24 -ac +extension GLX +render -noreset >"$XVFB_LOG" 2>&1 &
+XVFB_PID=$!
+PIDS+=("$XVFB_PID")
+log "started: Xvfb $DISPLAY (pid=$XVFB_PID)"
+
+if ! wait_for_x_display 15; then
+  log "fatal: Xvfb did not initialize DISPLAY=${DISPLAY}"
+  if [ -f "$XVFB_LOG" ]; then
+    log "xvfb log tail:"
+    tail -n 40 "$XVFB_LOG" || true
+  fi
+  exit 1
+fi
+
+start_bg fluxbox
+start_bg x11vnc -display "$DISPLAY" -rfbport "$VNC_PORT" -nopw -listen 0.0.0.0 -xkb -forever -shared
+start_bg websockify --web=/usr/share/novnc/ "$NOVNC_PORT" "localhost:${VNC_PORT}"
+start_bg socat "TCP-LISTEN:${CHROME_PUBLIC_DEBUG_PORT},fork,reuseaddr" "TCP:127.0.0.1:${CHROME_LOCAL_DEBUG_PORT}"
+start_bg python3 -u /src/browser_use_runner.py
+
+if ! wait_for_port 127.0.0.1 "$VNC_PORT" 20; then
+  log "fatal: x11vnc did not open port ${VNC_PORT}"
+  exit 1
+fi
+if ! wait_for_port 127.0.0.1 "$NOVNC_PORT" 20; then
+  log "fatal: websockify did not open port ${NOVNC_PORT}"
+  exit 1
+fi
+if ! wait_for_port 127.0.0.1 "$BROWSER_USE_RPC_PORT" 20; then
+  log "fatal: browser-use RPC did not open port ${BROWSER_USE_RPC_PORT}"
+  exit 1
+fi
+
+log "browser infrastructure is ready (noVNC:${NOVNC_PORT}, DevTools proxy:${CHROME_PUBLIC_DEBUG_PORT}, browser-use RPC:${BROWSER_USE_RPC_PORT})"
 
 while true; do
-  rm -f /src/browser_data/SingletonLock
+  for pid in "${PIDS[@]}"; do
+    if ! kill -0 "$pid" >/dev/null 2>&1; then
+      log "fatal: required background process died (pid=${pid})"
+      exit 1
+    fi
+  done
 
+  rm -f "${CHROME_PROFILE_DIR}/SingletonLock" "${CHROME_PROFILE_DIR}/SingletonCookie" "${CHROME_PROFILE_DIR}/SingletonSocket" 2>/dev/null || true
+
+  log "starting Chromium (local DevTools:${CHROME_LOCAL_DEBUG_PORT})"
   chromium \
     --no-sandbox \
     --disable-dev-shm-usage \
-    --remote-debugging-port=9223 \
+    --ozone-platform=x11 \
+    --remote-debugging-port="${CHROME_LOCAL_DEBUG_PORT}" \
     --remote-debugging-address=127.0.0.1 \
-    --remote-allow-origins=* \
+    --remote-allow-origins='*' \
     --window-size=1280,720 \
-    --user-data-dir=/src/browser_data \
+    --user-data-dir="${CHROME_PROFILE_DIR}" \
     --disable-blink-features=AutomationControlled \
     --no-first-run \
     --disable-gpu \
     --mute-audio \
     --no-default-browser-check \
     --disable-software-rasterizer \
-    --disable-features=site-per-process
+    --disable-features=site-per-process \
+    --disable-crash-reporter \
+    --disable-extensions \
+    --disable-sync &
 
-  echo "Chromium упал или был закрыт агентом, рестарт через 2 секунды..."
+  CHROME_PID=$!
-  sleep 2
+  wait "$CHROME_PID" || CHROME_EXIT=$?
+  CHROME_EXIT=${CHROME_EXIT:-0}
+
+  if [ "$STOPPING" -eq 1 ]; then
+    break
+  fi
+
+  now="$(date +%s)"
+  if [ $(( now - WINDOW_START )) -gt "$RESTART_WINDOW_SEC" ]; then
+    WINDOW_START="$now"
+    RESTART_COUNT=0
+  fi
+
+  RESTART_COUNT=$((RESTART_COUNT + 1))
+  log "Chromium exited with code=${CHROME_EXIT}; restart ${RESTART_COUNT}/${MAX_RESTARTS} in current window"
+
+  if [ "$RESTART_COUNT" -ge "$MAX_RESTARTS" ]; then
+    log "fatal: too many Chromium restarts in ${RESTART_WINDOW_SEC}s"
+    exit 1
+  fi
+
+  sleep "$RESTART_BACKOFF_SEC"
+  unset CHROME_EXIT
+  unset CHROME_PID
 done
+

docker-compose.yml

@@ -4,14 +4,11 @@ services:
       context: ./hermes_code
       dockerfile: Dockerfile
     container_name: hermes-brain
-    sysctls:
-      - net.ipv4.tcp_keepalive_time=60
-      - net.ipv4.tcp_keepalive_intvl=10
-      - net.ipv4.tcp_keepalive_probes=3
     env_file:
       - .env
     environment:
       - BROWSER_URL=http://browser:9222
+      - BROWSER_USE_RPC_URL=http://browser:8787/run
       - HERMES_HOME=/app/hermes_data
     volumes:
       - ./hermes_code:/app/hermes_code:ro
@@ -24,12 +21,6 @@ services:
     stdin_open: true
     tty: true
     restart: always
-    healthcheck:
-      test: ["CMD-SHELL", "pgrep -f 'python -m gateway.run' || exit 1"]
-      interval: 2m
-      timeout: 10s
-      retries: 3
-      start_period: 1m
     networks:
       - hermes-net
     deploy:
@@ -44,12 +35,13 @@ services:
       fi;
       exec python -m gateway.run
       "
-
   browser:
     build:
       context: ./browser_env
       dockerfile: Dockerfile.browser
     container_name: hermes-browser
+    env_file:
+      - .env
     ports:
       - "6080:6080"
       - "9222:9222"
@@ -62,14 +54,35 @@ services:
       - browser_profiles:/src/browser_data
     restart: always
     healthcheck:
-      test: ["CMD", "curl", "-f", "http://localhost:9222/json/version"]
+      test: [ "CMD-SHELL", "curl -fsS http://127.0.0.1:9222/json/version >/dev/null && curl -fsS http://127.0.0.1:8787/health >/dev/null || exit 1" ]
       interval: 10s
-      timeout: 5s
+      timeout: 3s
-      retries: 5
+      retries: 12
-      start_period: 10s
+      start_period: 20s
+
+  browser-api:
+    build:
+      context: ./api
+      dockerfile: Dockerfile
+    container_name: hermes-browser-api
+    environment:
+      - BROWSER_USE_RPC_URL=http://browser:8787/run
+      - BROWSER_API_HOST=0.0.0.0
+      - BROWSER_API_PORT=8088
+      - BROWSER_API_MAX_CONCURRENCY=2
+    depends_on:
+      browser:
+        condition: service_healthy
+    ports:
+      - "8088:8088"
+    restart: always
+    networks:
+      - hermes-net
 
   tunnel:
     image: cloudflare/cloudflared:latest
+    profiles:
+      - remote
     container_name: hermes-tunnel
     restart: always
     command: tunnel --protocol http2 --url http://browser:6080 --no-tls-verify

hermes_code/package.json

@@ -4,7 +4,7 @@
   "description": "An AI agent with advanced tool-calling capabilities, featuring a flexible toolsets system for organizing and managing tools.",
   "private": true,
   "scripts": {
-    "postinstall": "echo '✅ Browser tools ready. Run: python run_agent.py --help'"
+    "postinstall": "echo '✅ Hermes Node hooks ready. Run: python run_agent.py --help'"
   },
   "repository": {
     "type": "git",
@@ -15,9 +15,7 @@
     "url": "https://github.com/NousResearch/Hermes-Agent/issues"
   },
   "homepage": "https://github.com/NousResearch/Hermes-Agent#readme",
-  "dependencies": {
+  "dependencies": {},
-    "agent-browser": "^0.13.0"
-  },
   "engines": {
     "node": ">=18.0.0"
   }

hermes_code/pyproject.toml

@@ -34,10 +34,6 @@ dependencies = [
   "faster-whisper>=1.0.0,<2",
   # Skills Hub (GitHub App JWT auth — optional, only needed for bot identity)
   "PyJWT[crypto]>=2.10.1,<3",
-  "browser-use>=0.12.5",
-  "playwright>=1.49.0",
-  "playwright-stealth>=1.0.6",
-  "langchain-openai>=1.1.12",
 ]
 
 [project.optional-dependencies]

hermes_code/requirements.txt

@@ -19,9 +19,6 @@ PyJWT[crypto]
 # Web tools
 firecrawl-py
 parallel-web>=0.4.2
-browser-use>=0.12.5
-playwright
-playwright-stealth
 
 # Image generation
 fal-client
@@ -36,5 +33,4 @@ croniter
 python-telegram-bot>=20.0
 discord.py>=2.0
 aiohttp>=3.9.0
-langchain-openai>=1.1.12,
 

hermes_code/scripts/install.sh

@@ -874,41 +874,14 @@ install_node_deps() {
     fi
 
     if [ -f "$INSTALL_DIR/package.json" ]; then
-        log_info "Installing Node.js dependencies (browser tools)..."
+        log_info "Installing Node.js dependencies..."
         cd "$INSTALL_DIR"
         npm install --silent 2>/dev/null || {
-            log_warn "npm install failed (browser tools may not work)"
+            log_warn "npm install failed (some Node.js helpers may not work)"
         }
         log_success "Node.js dependencies installed"
 
-        # Install Playwright browser + system dependencies.
+        log_info "Skipping local Playwright/Chromium bootstrap (browser automation runs in browser container)"
-        # Playwright's install-deps only supports apt/dnf/zypper natively.
-        # For Arch/Manjaro we install the system libs via pacman first.
-        log_info "Installing browser engine (Playwright Chromium)..."
-        case "$DISTRO" in
-            arch|manjaro)
-                if command -v pacman &> /dev/null; then
-                    log_info "Arch/Manjaro detected — installing Chromium system dependencies via pacman..."
-                    if command -v sudo &> /dev/null && sudo -n true 2>/dev/null; then
-                        sudo NEEDRESTART_MODE=a pacman -S --noconfirm --needed \
-                            nss atk at-spi2-core cups libdrm libxkbcommon mesa pango cairo alsa-lib >/dev/null 2>&1 || true
-                    elif [ "$(id -u)" -eq 0 ]; then
-                        pacman -S --noconfirm --needed \
-                            nss atk at-spi2-core cups libdrm libxkbcommon mesa pango cairo alsa-lib >/dev/null 2>&1 || true
-                    else
-                        log_warn "Cannot install browser deps without sudo. Run manually:"
-                        log_warn "  sudo pacman -S nss atk at-spi2-core cups libdrm libxkbcommon mesa pango cairo alsa-lib"
-                    fi
-                fi
-                cd "$INSTALL_DIR" && npx playwright install chromium 2>/dev/null || true
-                ;;
-            *)
-                log_info "Playwright may request sudo to install browser system dependencies (shared libraries)."
-                log_info "This is standard Playwright setup — Hermes itself does not require root access."
-                cd "$INSTALL_DIR" && npx playwright install --with-deps chromium 2>/dev/null || true
-                ;;
-        esac
-        log_success "Browser engine installed"
     fi
 
     # Install WhatsApp bridge dependencies

hermes_code/tools/browser_use_tool.py

@@ -1,70 +1,40 @@
 import json
 import os
-import asyncio
+from urllib import error, request
-import socket
-from browser_use import Agent, Browser, ChatOpenAI
 from tools.registry import registry
 
 
-async def run_browser_task(task):
+def run_browser_task(task):
-    browser_host = "browser"
+    if not task or not str(task).strip():
-    browser_port = 9222
+        return json.dumps({"success": False, "error": "Task is required"}, ensure_ascii=False)
-    BROWSER_VIEW_URL = os.getenv("BROWSER_VIEW_URL", "")
+
+    rpc_url = os.getenv("BROWSER_USE_RPC_URL", "http://browser:8787/run")
+    timeout_sec = int(os.getenv("BROWSER_USE_RPC_TIMEOUT", "900"))
+    payload = json.dumps({"task": task}).encode("utf-8")
+    req = request.Request(rpc_url, data=payload, headers={"Content-Type": "application/json"}, method="POST")
 
     try:
-        browser_ip = socket.gethostbyname(browser_host)
+        with request.urlopen(req, timeout=timeout_sec) as resp:
-        cdp_url = f"http://{browser_ip}:{browser_port}"
+            body = resp.read().decode("utf-8")
-    except Exception:
+            return body
-        cdp_url = f"http://{browser_host}:{browser_port}"
+    except error.HTTPError as http_err:
-
+        body = http_err.read().decode("utf-8", errors="replace")
-    browser = Browser(cdp_url=cdp_url)
+        return json.dumps(
-
+            {
-    # Для подключения к Chrome на виртуальной машине раскомментируй
+                "success": False,
-    # browser = Browser(
+                "error": f"browser-use RPC returned HTTP {http_err.code}",
-    #     executable_path="/usr/bin/google-chrome",  # Linux
+                "details": body,
-    #     # Windows: "C:\\Program Files\\Google\\Chrome\\Application\\chrome.exe"
+            },
-    #     # macOS: "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"
+            ensure_ascii=False,
-    # )
+        )
-    # или
+    except Exception as err:
-    # browser = Browser.from_system_chrome() для автоопределения
+        return json.dumps(
-
+            {
-    llm = ChatOpenAI(
+                "success": False,
-        model=os.getenv("MODEL_DEFAULT", "qwen3.5-122b"),
+                "error": f"browser-use RPC request failed: {err}",
-        api_key=os.getenv("OPENAI_API_KEY"),
+            },
-        base_url=os.getenv("OPENAI_BASE_URL"),
+            ensure_ascii=False,
-        temperature=0.0,
+        )
-    )
-
-    agent = Agent(
-        task=task,
-        llm=llm,
-        browser=browser,
-        use_vision=False
-    )
-
-    try:
-        history = await agent.run()
-        final_result = history.final_result()
-
-        response = {
-            "success": True,
-            "result": final_result,
-            "browser_view": BROWSER_VIEW_URL
-        }
-        return json.dumps(response, ensure_ascii=False)
-        
-    except Exception as e:
-        return json.dumps({
-            "success": False, 
-            "error": f"Browser automation failed: {str(e)}"
-        }, ensure_ascii=False)
-        
-    finally:
-        if browser:
-            try:
-                await browser.close()
-            except Exception:
-                pass
 
 
 registry.register(
@@ -90,6 +60,6 @@ registry.register(
         }
     },
  
-    handler=lambda args, **kw: asyncio.run(run_browser_task(args.get("task"))),
+    handler=lambda args, **kw: run_browser_task(args.get("task")),
     emoji="🌐",
 )