Windows Support

README.md

@@ -17,7 +17,7 @@
 
 **Fara-7B** is Microsoft's first **agentic small language model (SLM)** designed specifically for computer use. With only 7 billion parameters, Fara-7B is an ultra-compact Computer Use Agent (CUA) that achieves state-of-the-art performance within its size class and is competitive with larger, more resource-intensive agentic systems.
 
-Try Fara-7B locally as follows (see [Installation](##Installation) for detailed instructions) or via Magentic-UI:
+Try Fara-7B locally as follows (see [Installation](#Installation) for detailed instructions on Windows ) or via Magentic-UI:
 
 ```bash
 # 1. Clone repository
@@ -44,7 +44,7 @@ To try Fara-7B inside Magentic-UI, please follow the instructions here [Magentic
 
 
 Notes:
-- If you're using Windows, we highly recommend using WSL2 (Windows Subsystem for Linux).
+- If you're using Windows, we highly recommend using WSL2 (Windows Subsystem for Linux). Please the Windows instructions in the [Installation](#Installation) section.
 - You might need to do `--tensor-parallel-size 2` with vllm command if you run out of memory
 
 <table>
@@ -156,27 +156,45 @@ Our evaluation setup leverages:
 
 ---
 
-## Installation
+# Installation
 
-Install the package using either UV or pip:
 
-```bash
-uv sync --all-extras
-```
+##  Linux 
+
+The following instructions are for Linux systems, see the Windows section below for Windows instructions. 
 
-or
+Install the package using pip and set up the environment with Playwright:
 
 ```bash
-pip install -e .
+# 1. Clone repository
+git clone https://github.com/microsoft/fara.git
+cd fara
+
+# 2. Setup environment
+python3 -m venv .venv 
+source .venv/bin/activate
+pip install -e .[vllm]
+playwright install
 ```
 
-Then install Playwright browsers:
+Note: If you plan on hosting with Azure Foundry only, you can skip the `[vllm]` and just do `pip install -e .`
+
+
+## Windows
+
+For Windows, we highly recommend using WSL2 (Windows Subsystem for Linux) to provide a Linux-like environment. However, if you prefer to run natively on Windows, follow these steps:
 
 ```bash
-playwright install
-```
+# 1. Clone repository
+git clone https://github.com/microsoft/fara.git
+cd fara
 
----
+# 2. Setup environment
+python3 -m venv .venv
+.venv\Scripts\activate
+pip install -e .
+python3 -m playwright install
+```
 
 ## Hosting the Model
 
@@ -189,74 +207,66 @@ Deploy Fara-7B on [Azure Foundry](https://ai.azure.com/explore/models/Fara-7B/ve
 **Setup:**
 
 1. Deploy the Fara-7B model on Azure Foundry and obtain your endpoint URL and API key
-2. Add your endpoint details to the existing `endpoint_configs/` directory (example configs are already provided):
 
-```bash
-# Edit one of the existing config files or create a new one
-# endpoint_configs/fara-7b-hosting-ansrz.json (example format):
+Then create a endpoint configuration JSON file (e.g., `azure_foundry_config.json`):
+
+```json
 {
     "model": "Fara-7B",
     "base_url": "https://your-endpoint.inference.ml.azure.com/",
     "api_key": "YOUR_API_KEY_HERE"
 }
 ```
 
-3. Run the Fara agent:
+Then you can run Fara-7B using this endpoint configuration.
+
+2. Run the Fara agent:
 
 ```bash
-fara-cli --task "how many pages does wikipedia have" --start_page "https://www.bing.com"
+fara-cli --task "how many pages does wikipedia have" --endpoint_config azure_foundry_config.json [--headful]
 ```
 
-That's it! No GPU or model downloads required.
+Note: you can also specify the endpoint config with the args `--base_url [your_base_url] --api_key [your_api_key] --model [your_model_name]` instead of using a config JSON file. 
 
-### Self-hosting with VLLM
-
-If you have access to GPU resources, you can self-host Fara-7B using VLLM. This requires a GPU machine with sufficient VRAM.
-
-All that is required is to run the following command to start the VLLM server:
+Note: If you see an error that the `fara-cli` command is not found, then try:
 
 ```bash
-vllm serve "microsoft/Fara-7B" --port 5000 --dtype auto 
+python -m fara.run_fara --task "what is the weather in new york now"
 ```
 
-### Testing the Fara Agent
+That's it! No GPU or model downloads required.
 
-Run the test script to see Fara in action:
+### Self-hosting with vLLM or LM Studio / Ollama
+
+**If you have access to GPU resources, you can self-host Fara-7B using vLLM. This requires a GPU machine with sufficient VRAM (e.g., 24GB or more).**
+
+Only on Linux: all that is required is to run the following command to start the VLLM server:
 
 ```bash
-fara-cli --task "how many pages does wikipedia have" --start_page "https://www.bing.com" --endpoint_config endpoint_configs/azure_foundry_config.json [--headful] [--downloads_folder "/path/to/downloads"] [--save_screenshots] [--max_rounds 100] [--browserbase]
+vllm serve "microsoft/Fara-7B" --port 5000 --dtype auto 
 ```
+For quantized models or lower VRAM GPUs, please see [Fara-7B GGUF on HuggingFace](https://huggingface.co/bartowski/microsoft_Fara-7B-GGUF).
 
-In self-hosting scenario the `endpoint_config` points to `endpoint_configs/vllm_config.json` from the VLLM server above.
+** For Windows/Mac, vLLM is not natively supported. You can use WSL2 on Windows to run the above command or LM Studio / Ollama as described below. **
 
-If you set `--browserbase`, export environment variables for the API key and project ID.
+Otherwise, you can use [LM Studio](https://lmstudio.ai/) or [Ollama](https://ollama.com/) to host the model locally. We currently recommend the following GGUF versions of our models [Fara-7B GGUF on HuggingFace](https://huggingface.co/bartowski/microsoft_Fara-7B-GGUF) for use with LM Studio or Ollama. Select the largest model that fits your GPU. Please ensure that context length is set to at least 15000 tokens and temperature to 0 for best results.
 
-#### Expected Output
-
-```
-Initializing Browser...
-Browser Running... Starting Fara Agent...
-##########################################
-Task: how many pages does wikipedia have
-##########################################
-Running Fara...
+Then you can run Fara-7B pointing to your local  server:
 
+Run the test script to see Fara in action:
 
-Thought #1: To find the current number of Wikipedia pages, I'll search for the latest Wikipedia page count statistics.
-Action #1: executing tool 'web_search' with arguments {"action": "web_search", "query": "Wikipedia total number of articles"}
-Observation#1: I typed 'Wikipedia total number of articles' into the browser search bar.
+```bash
+fara-cli --task "what is the weather in new york now"
+```
 
-Thought #2: Wikipedia currently has 7,095,446 articles.
-Action #2: executing tool 'terminate' with arguments {"action": "terminate", "status": "success"}
-Observation#2: Wikipedia currently has 7,095,446 articles.
+If you didn't use vLLM to host, please specify the correct `--base_url [your_base_url] --api_key [your_api_key] --model [your_model_name]`
 
-Final Answer: Wikipedia currently has 7,095,446 articles.
+If you see an error that the `fara-cli` command is not found, then try:
 
-Enter another task (or press Enter to exit): 
+```bash
+python -m fara.run_fara --task "what is the weather in new york now"
 ```
 
----
-
 # Reproducibility
 
 We provide a framework in `webeval/` to reproduce our results on WebVoyager and OnlineMind2Web. 

pyproject.toml

@@ -32,14 +32,20 @@ dependencies = [
     "pyyaml",
     "jsonschema",
     "browserbase",
-    "vllm>=0.10.0"
 ]
 
+
+
+
 [project.urls]
 Homepage = "https://github.com/microsoft/fara"
 Repository = "https://github.com/microsoft/fara"
 Issues = "https://github.com/microsoft/fara/issues"
 
+[project.optional-dependencies]
+vllm = ["vllm>=0.10.0"]
+lmstudio = ["lmstudio"]
+ollama = ["ollama"]
 
 [project.scripts]
 fara-cli = "fara.run_fara:main"

src/fara/browser/browser_bb.py

@@ -6,7 +6,7 @@
 import subprocess
 import time
 from typing import Any, Dict, Optional, Callable
-
+import platform
 import browserbase
 from browserbase import Browserbase
 from playwright.async_api import (
@@ -48,7 +48,7 @@ def __init__(
         self.single_tab_mode = single_tab_mode
         self.use_browser_base = use_browser_base
         self.logger = logger or logging.getLogger("browser_manager")
-
+        self.is_linux = platform.system() == "Linux"
         self._viewport_height = viewport_height
         self._viewport_width = viewport_width
 
@@ -194,7 +194,8 @@ async def delayed_resume():
 
     async def _init_regular_browser(self, channel: str = "chromium") -> None:
         """Initialize regular browser according to the specified channel."""
-        if not self.headless:
+        if not self.headless and self.is_linux:
+            print("STARTING XVFB")
             self.start_xvfb()
 
         launch_args: Dict[str, Any] = {"headless": self.headless}
@@ -218,7 +219,7 @@ async def _init_regular_browser(self, channel: str = "chromium") -> None:
 
     async def _init_persistent_browser(self) -> None:
         """Initialize persistent browser with data directory."""
-        if not self.headless:
+        if not self.headless and self.is_linux:
             self.start_xvfb()
 
         launch_args: Dict[str, Any] = {"headless": self.headless}

src/fara/fara_agent.py

@@ -15,7 +15,7 @@
 import asyncio
 from .browser.playwright_controller import PlaywrightController
 from ._prompts import get_computer_use_system_prompt
-from .types import (
+from .fara_types import (
     LLMMessage,
     SystemMessage,
     UserMessage,
@@ -379,15 +379,20 @@ async def run(self, user_message: str) -> Tuple:
             thoughts, action_dict = self._parse_thoughts_and_action(raw_response)
             action_args = action_dict.get("arguments", {})
             action = action_args["action"]
-            self.logger.info(f"\nThought #{i+1}: {thoughts}\nAction #{i+1}: executing tool '{action}' with arguments {json.dumps(action_args)}")
-
+            self.logger.debug(
+                f"\nThought #{i+1}: {thoughts}\nAction #{i+1}: executing tool '{action}' with arguments {json.dumps(action_args)}"
+            )
+            print(
+                f"\nThought #{i+1}: {thoughts}\nAction #{i+1}: executing tool '{action}' with arguments {json.dumps(action_args)}"
+            )
             (
                 is_stop_action,
                 new_screenshot,
                 action_description,
             ) = await self.execute_action(function_call)
             all_observations.append(action_description)
-            self.logger.info(f"Observation#{i+1}: {action_description}")
+            self.logger.debug(f"Observation#{i+1}: {action_description}")
+            print(f"Observation#{i+1}: {action_description}")
             if is_stop_action:
                 final_answer = thoughts
                 break
@@ -564,7 +569,7 @@ async def execute_action(
         elif args["action"] == "pause_and_memorize_fact":
             fact = str(args.get("fact"))
             self._facts.append(fact)
-            action_description= f"I memorized the following fact: {fact}"
+            action_description = f"I memorized the following fact: {fact}"
         elif args["action"] == "stop" or args["action"] == "terminate":
             action_description = args.get("thoughts")
             is_stop_action = True

src/fara/run_fara.py

@@ -1,8 +1,8 @@
 import asyncio
 import argparse
 import os
-from fara import FaraAgent
-from fara.browser.browser_bb import BrowserBB
+from .fara_agent import FaraAgent
+from .browser.browser_bb import BrowserBB
 import logging
 from typing import Dict
 from pathlib import Path
@@ -11,8 +11,8 @@
 
 # Configure logging to only show logs from fara.fara_agent
 logging.basicConfig(
-    level=logging.CRITICAL,  
-    format="%(message)s",  
+    level=logging.CRITICAL,
+    format="%(message)s",
 )
 
 # Enable INFO level only for fara.fara_agent
@@ -159,21 +159,51 @@ def main():
         default=None,
         help="Path to the endpoint configuration JSON file. By default, tries local vllm on 5000 port",
     )
+    parser.add_argument(
+        "--api_key",
+        type=str,
+        default=None,
+        help="API key for the model endpoint (overrides endpoint_config)",
+    )
+    parser.add_argument(
+        "--base_url",
+        type=str,
+        default=None,
+        help="Base URL for the model endpoint (overrides endpoint_config)",
+    )
+    parser.add_argument(
+        "--model",
+        type=str,
+        default=None,
+        help="Model name to use (overrides endpoint_config)",
+    )
 
     args = parser.parse_args()
 
     if args.browserbase:
-        assert os.environ.get("BROWSERBASE_API_KEY"), (
-            "BROWSERBASE_API_KEY environment variable must be set to use browserbase"
-        )
-        assert os.environ.get("BROWSERBASE_PROJECT_ID"), (
-            "BROWSERBASE_API_KEY and BROWSERBASE_PROJECT_ID environment variables must be set to use browserbase"
-        )
+        assert os.environ.get(
+            "BROWSERBASE_API_KEY"
+        ), "BROWSERBASE_API_KEY environment variable must be set to use browserbase"
+        assert os.environ.get(
+            "BROWSERBASE_PROJECT_ID"
+        ), "BROWSERBASE_API_KEY and BROWSERBASE_PROJECT_ID environment variables must be set to use browserbase"
 
     endpoint_config = DEFAULT_ENDPOINT_CONFIG
     if args.endpoint_config:
         with open(args.endpoint_config, "r") as f:
             endpoint_config = json.load(f)
+            assert (
+                "api_key" in endpoint_config
+                and "base_url" in endpoint_config
+                and "model" in endpoint_config
+            ), "endpoint_config file must contain api_key, base_url, and model fields"
+    # Override with command-line arguments if provided
+    if args.api_key:
+        endpoint_config["api_key"] = args.api_key
+    if args.base_url:
+        endpoint_config["base_url"] = args.base_url
+    if args.model:
+        endpoint_config["model"] = args.model
 
     asyncio.run(
         run_fara_agent(