require tokens to be input by user in the UI before enabling the classify button.

app.py

@@ -15,17 +15,14 @@ from tavily import TavilyClient
 from huggingface_hub import InferenceClient
 
 text_classifier = None
-TAVILY_KEY = os.getenv("TAVILY_API_KEY")
-GOOGLE_KEY = os.getenv("GOOGLE_FC_KEY")
-HF_TOKEN = os.getenv("HF_TOKEN")
+TAVILY_KEY = None
+GOOGLE_KEY = None
+HF_TOKEN = None
 
 embed_model = SentenceTransformer("all-MiniLM-L6-v2")
 explain_model = "meta-llama/Llama-3.1-8B-Instruct"
 text_model = "rajyalakshmijampani/fever_finetuned_deberta"
-
-inf_client = InferenceClient(token=HF_TOKEN)
 wiki = wikipediaapi.Wikipedia(language='en', user_agent='fact-checker/1.0')
-tavily = TavilyClient(api_key=TAVILY_KEY)
 
 def get_text_classifier():
     global text_classifier
@@ -66,7 +63,7 @@ def _safe_call(func, claim):
         return []
 
 def _from_google(claim):
-    if not GOOGLE_KEY: return []
+    global GOOGLE_KEY
     url = "https://factchecktools.googleapis.com/v1alpha1/claims:search"
     r = requests.get(url, params={"query": claim, "key": GOOGLE_KEY, "pageSize": 2}).json()
     claims = r.get("claims", [])
@@ -80,7 +77,8 @@ def _from_google(claim):
     return evid[:3]
 
 def _from_tavily(claim):
-    if not TAVILY_KEY: return []
+    global TAVILY_KEY
+    tavily = TavilyClient(api_key=TAVILY_KEY)
     try:
         results = tavily.search(claim).get("results", [])
         sents = []
@@ -118,7 +116,13 @@ def get_evidence_sentences(claim, k=3):
     return (evid or ["Error: No relevant evidence found."])[:k]
 
 # --- Classification Function ---
-def classify_text(claim):
+def classify_text(claim, hf_token, tavily_key, google_key):
+
+    global HF_TOKEN, TAVILY_KEY, GOOGLE_KEY
+    HF_TOKEN = hf_token.strip()
+    TAVILY_KEY = tavily_key.strip()
+    GOOGLE_KEY = google_key.strip()
+
     claim=claim.lower().strip()
     classifier = get_text_classifier()
     evidences = get_evidence_sentences(claim)    
@@ -159,6 +163,8 @@ def classify_text(claim):
         {"role": "system", "content": "You are a reliable fact-checking assistant."},
         {"role": "user", "content": prompt},
     ]
+
+    inf_client = InferenceClient(token=HF_TOKEN)
     completion = inf_client.chat_completion( model=explain_model, messages=messages, max_tokens=256, temperature=0.3)
     raw_response = completion.choices[0].message.content.strip()
 
@@ -204,10 +210,32 @@ with gr.Blocks() as demo:
     gr.Markdown("# Multimodal Misinformation Detector")
 
     with gr.Tab("Text Detector"):
-        claim = gr.Textbox(label="Enter Claim")
-        text_button = gr.Button("Classify Claim")
-        text_output = gr.Markdown(label="Model Output", value="Results will appear here...",)
-        text_button.click(classify_text, inputs=claim, outputs=text_output)
+        with gr.Row(): 
+            with gr.Column(scale=3): #  Left half — main inputs
+                claim = gr.Textbox(label="Enter Claim")
+                text_button = gr.Button("Classify Claim", interactive=False) # Disable until tokens provided
+                text_output = gr.Markdown( label="Model Output", value="Results will appear here...")
+
+            
+            with gr.Column(scale=1):  # Right half — user token inputs
+                gr.Markdown("## Enter your API keys")
+                hf_token = gr.Textbox(label="Hugging Face Token", type="password", value = "Required")
+                tavily_key = gr.Textbox(label="Tavily API Key", type="password", value = "Required")
+                google_key = gr.Textbox(label="Google Fact Check API Key", type="password", value = "Required")
+            
+        # Enable button when all fields filled
+        def enable_button(hf, tavily, google):
+            ready = bool(hf and tavily and google)
+            return gr.update(interactive=ready)
+        
+        hf_token.change(enable_button, inputs=[hf_token, tavily_key, google_key], outputs=text_button)
+        tavily_key.change(enable_button, inputs=[hf_token, tavily_key, google_key], outputs=text_button)
+        google_key.change(enable_button, inputs=[hf_token, tavily_key, google_key], outputs=text_button)
+
+        # Click handler (include all token inputs)
+        text_button.click(classify_text,
+                          inputs=[claim, hf_token, tavily_key, google_key],
+                          outputs=text_output)
         
 
     with gr.Tab("Image Detector"):