diff --git a/README.md b/README.md
index 58ca928..f4daeea 100644
--- a/README.md
+++ b/README.md
@@ -77,13 +77,7 @@ This is a tutorial project of [Pocket Flow](https://github.com/The-Pocket/Pocket
    pip install -r requirements.txt
    ```
 
-3. Set up LLM in [`utils/call_llm.py`](./utils/call_llm.py) by providing credentials. By default, you can use the AI Studio key with this client for Gemini Pro 2.5:
-
-   ```python
-   client = genai.Client(
-     api_key=os.getenv("GEMINI_API_KEY", "your-api_key"),
-   )
-   ```
+3. Set up LLM in [`utils/call_llm.py`](./utils/call_llm.py) by providing credentials. To do so, you can put the values in a `.env` file. By default, you can use the AI Studio key with this client for Gemini Pro 2.5 by setting the `GEMINI_API_KEY` environment variable. If you want to use another LLM, you can set the `LLM_PROVIDER` environment variable (e.g. `XAI`), and then set the model, url, and API key (e.g. `XAI_MODEL`, `XAI_URL`,`XAI_API_KEY`). If using Ollama, the url is `http://localhost:11434/` and the API key can be omitted.
 
    You can use your own models. We highly recommend the latest models with thinking capabilities (Claude 3.7 with thinking, O1). You can verify that it is correctly set up by running:
    ```bash
diff --git a/utils/call_llm.py b/utils/call_llm.py
index 0d794b4..9f21a44 100644
--- a/utils/call_llm.py
+++ b/utils/call_llm.py
@@ -2,6 +2,7 @@
 import os
 import logging
 import json
+import requests
 from datetime import datetime
 
 # Configure logging
@@ -20,69 +21,158 @@
 # Simple cache configuration
 cache_file = "llm_cache.json"
 
+
+def load_cache():
+    try:
+        with open(cache_file, 'r') as f:
+            return json.load(f)
+    except:
+        logger.warning(f"Failed to load cache.")
+    return {}
+
+
+def save_cache(cache):
+    try:
+        with open(cache_file, 'w') as f:
+            json.dump(cache, f)
+    except:
+        logger.warning(f"Failed to save cache")
+
+
+def get_llm_provider():
+    provider = os.getenv("LLM_PROVIDER")
+    if not provider and (os.getenv("GEMINI_PROJECT_ID") or os.getenv("GEMINI_API_KEY")):
+        provider = "GEMINI"
+    # if necessary, add ANTHROPIC/OPENAI
+    return provider
+
+
+def _call_llm_provider(prompt: str) -> str:
+    """
+    Call an LLM provider based on environment variables.
+    Environment variables:
+    - LLM_PROVIDER: "OLLAMA" or "XAI"
+    - <provider>_MODEL: Model name (e.g., OLLAMA_MODEL, XAI_MODEL)
+    - <provider>_BASE_URL: Base URL without endpoint (e.g., OLLAMA_BASE_URL, XAI_BASE_URL)
+    - <provider>_API_KEY: API key (e.g., OLLAMA_API_KEY, XAI_API_KEY; optional for providers that don't require it)
+    The endpoint /v1/chat/completions will be appended to the base URL.
+    """
+    logger.info(f"PROMPT: {prompt}") # log the prompt
+
+    # Read the provider from environment variable
+    provider = os.environ.get("LLM_PROVIDER")
+    if not provider:
+        raise ValueError("LLM_PROVIDER environment variable is required")
+
+    # Construct the names of the other environment variables
+    model_var = f"{provider}_MODEL"
+    base_url_var = f"{provider}_BASE_URL"
+    api_key_var = f"{provider}_API_KEY"
+
+    # Read the provider-specific variables
+    model = os.environ.get(model_var)
+    base_url = os.environ.get(base_url_var)
+    api_key = os.environ.get(api_key_var, "")  # API key is optional, default to empty string
+
+    # Validate required variables
+    if not model:
+        raise ValueError(f"{model_var} environment variable is required")
+    if not base_url:
+        raise ValueError(f"{base_url_var} environment variable is required")
+
+    # Append the endpoint to the base URL
+    url = f"{base_url.rstrip('/')}/v1/chat/completions"
+
+    # Configure headers and payload based on provider
+    headers = {
+        "Content-Type": "application/json",
+    }
+    if api_key:  # Only add Authorization header if API key is provided
+        headers["Authorization"] = f"Bearer {api_key}"
+
+    payload = {
+        "model": model,
+        "messages": [{"role": "user", "content": prompt}],
+        "temperature": 0.7,
+    }
+
+    try:
+        response = requests.post(url, headers=headers, json=payload)
+        response_json = response.json() # Log the response
+        logger.info("RESPONSE:\n%s", json.dumps(response_json, indent=2))
+        #logger.info(f"RESPONSE: {response.json()}")
+        response.raise_for_status()
+        return response.json()["choices"][0]["message"]["content"]
+    except requests.exceptions.HTTPError as e:
+        error_message = f"HTTP error occurred: {e}"
+        try:
+            error_details = response.json().get("error", "No additional details")
+            error_message += f" (Details: {error_details})"
+        except:
+            pass
+        raise Exception(error_message)
+    except requests.exceptions.ConnectionError:
+        raise Exception(f"Failed to connect to {provider} API. Check your network connection.")
+    except requests.exceptions.Timeout:
+        raise Exception(f"Request to {provider} API timed out.")
+    except requests.exceptions.RequestException as e:
+        raise Exception(f"An error occurred while making the request to {provider}: {e}")
+    except ValueError:
+        raise Exception(f"Failed to parse response as JSON from {provider}. The server might have returned an invalid response.")
+
+
 # By default, we Google Gemini 2.5 pro, as it shows great performance for code understanding
 def call_llm(prompt: str, use_cache: bool = True) -> str:
     # Log the prompt
     logger.info(f"PROMPT: {prompt}")
-    
+
     # Check cache if enabled
     if use_cache:
         # Load cache from disk
-        cache = {}
-        if os.path.exists(cache_file):
-            try:
-                with open(cache_file, 'r') as f:
-                    cache = json.load(f)
-            except:
-                logger.warning(f"Failed to load cache, starting with empty cache")
-        
+        cache = load_cache()
         # Return from cache if exists
         if prompt in cache:
             logger.info(f"RESPONSE: {cache[prompt]}")
             return cache[prompt]
-    
-    # Call the LLM if not in cache or cache disabled
-    client = genai.Client(
-        vertexai=True, 
-        # TODO: change to your own project id and location
-        project=os.getenv("GEMINI_PROJECT_ID", "your-project-id"),
-        location=os.getenv("GEMINI_LOCATION", "us-central1")
-    )
-    # You can comment the previous line and use the AI Studio key instead:
-    # client = genai.Client(
-    #     api_key=os.getenv("GEMINI_API_KEY", "your-api_key"),
-    # )
-    model = os.getenv("GEMINI_MODEL", "gemini-2.5-pro-exp-03-25")
-    response = client.models.generate_content(
-        model=model,
-        contents=[prompt]
-    )
-    response_text = response.text
-    
+
+    provider = get_llm_provider()
+    if provider == "GEMINI":
+        response_text = _call_llm_gemini(prompt)
+    else:  # generic method using a URL that is OpenAI compatible API (Ollama, ...)
+        response_text = _call_llm_provider(prompt)
+
     # Log the response
     logger.info(f"RESPONSE: {response_text}")
-    
+
     # Update cache if enabled
     if use_cache:
         # Load cache again to avoid overwrites
-        cache = {}
-        if os.path.exists(cache_file):
-            try:
-                with open(cache_file, 'r') as f:
-                    cache = json.load(f)
-            except:
-                pass
-        
+        cache = load_cache()
         # Add to cache and save
         cache[prompt] = response_text
-        try:
-            with open(cache_file, 'w') as f:
-                json.dump(cache, f)
-        except Exception as e:
-            logger.error(f"Failed to save cache: {e}")
-    
+        save_cache(cache)
+
     return response_text
 
+
+def _call_llm_gemini(prompt: str) -> str:
+    if os.getenv("GEMINI_PROJECT_ID"):
+        client = genai.Client(
+            vertexai=True,
+            project=os.getenv("GEMINI_PROJECT_ID"),
+            location=os.getenv("GEMINI_LOCATION", "us-central1")
+        )
+    elif os.getenv("GEMINI_API_KEY"):
+        client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))
+    else:
+        raise ValueError("Either GEMINI_PROJECT_ID or GEMINI_API_KEY must be set in the environment")
+    model = os.getenv("GEMINI_MODEL", "gemini-2.5-pro-exp-03-25")
+    response = client.models.generate_content(
+        model=model,
+        contents=[prompt]
+    )
+    return response.text
+
 # # Use Anthropic Claude 3.7 Sonnet Extended Thinking
 # def call_llm(prompt, use_cache: bool = True):
 #     from anthropic import Anthropic
@@ -101,7 +191,7 @@ def call_llm(prompt: str, use_cache: bool = True) -> str:
 #     return response.content[1].text
 
 # # Use OpenAI o1
-# def call_llm(prompt, use_cache: bool = True):    
+# def call_llm(prompt, use_cache: bool = True):
 #     from openai import OpenAI
 #     client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY", "your-api-key"))
 #     r = client.chat.completions.create(
@@ -117,9 +207,9 @@ def call_llm(prompt: str, use_cache: bool = True) -> str:
 
 if __name__ == "__main__":
     test_prompt = "Hello, how are you?"
-    
+
     # First call - should hit the API
     print("Making call...")
     response1 = call_llm(test_prompt, use_cache=False)
     print(f"Response: {response1}")
-    
+