app.py

import gradio as gr
import requests
import os
import random
import pandas as pd

# Load instructions from local files
def load_instruction(persona):
    try:
        with open(f"instructions/{persona.lower()}.txt", "r") as file:
            return file.read()
    except FileNotFoundError:
        return ""

# Call Cohere API
def call_cohere_api(system_instruction, user_prompt):
    headers = {
        "Authorization": f"Bearer {os.getenv('COHERE_API_KEY')}",
        "Content-Type": "application/json"
    }
    
    # Append word limit instruction
    user_prompt += "\n\nWhen possible, make your answer relevant to Bristol or surrounding South West England context."
    user_prompt += "\n\nAnswer in 100 words or fewer."
    payload = {
        "model": "command-r-plus",
        "message": user_prompt,
        "preamble": system_instruction,
        "max_tokens": 300
    }
    response = requests.post("https://api.cohere.ai/v1/chat", headers=headers, json=payload)
    return response.json().get("text", "No response").strip()

# Load questions from file
def load_questions():
    try:
        with open("questions.txt", "r") as file:
            return [line.strip() for line in file if line.strip()]
    except FileNotFoundError:
        return []

questions_list = load_questions()

# Generate random question
def get_random_question():
    return random.choice(questions_list) if questions_list else "No questions available."

# Load counter-narratives CSV
def load_counternarratives():
    try:
        df = pd.read_csv("counternarratives.csv")
        return df
    except FileNotFoundError:
        print("counternarratives.csv not found.")
        return pd.DataFrame(columns=["myth", "fact", "persona"])

counternarratives = load_counternarratives()

# Generate Random Myth or Fact and trigger persona response
def get_random_myth_or_fact():
    if counternarratives.empty:
        return "No myths or facts available.", "Fact-Checker", "", "", ""

    # 🔄 Randomly select a row from the dataframe
    row = counternarratives.sample(1).iloc[0]
    selected_column = random.choice(["myth", "fact"])
    myth_or_fact = row[selected_column]
    persona = row["persona"]

    # 🔄 Call the Cohere API to get the persona's response
    persona_instruction = load_instruction(persona)
    persona_response = call_cohere_api(persona_instruction, myth_or_fact)

    # preparat the fact checker response       
    fact_check_response = f"{myth_or_fact} - Myth or Fact?\n\n"

    # ✅ Fact-checker response logic
    if selected_column == "myth":
        fact_check_response += f"❌ **MYTH**\nThe fact is: {row['fact']}"
    else:
        fact_response=call_cohere_api("You are an ecolinguistic aware assistant.", f"Elaborate on this fact: {row['fact']}")
        fact_check_response += f"✅ **FACT**\n{fact_response}"

    # Return the myth/fact, update the personas, and fill the responses
    return myth_or_fact, persona, persona, fact_check_response, persona_response,f"### Fact Checker", f"### .. and what the {persona} would say about it?:"

def ask_with_titles(p1, p2, q):
    # Generate responses
    response1 = call_cohere_api(load_instruction(p1), q)
    response2 = call_cohere_api(load_instruction(p2), q)

    # Generate titles
    title1 = f"### {p1} Responds"
    title2 = f"### {p2} Responds"
    
    # Return responses and titles
    return response1, response2, title1, title2

# Dynamically load persona names from instructions folder
personas = [os.path.splitext(f)[0].capitalize() for f in os.listdir("instructions") if f.endswith(".txt")]

# Gradio Interface
with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column(scale=0.10):
            gr.Image(value="data/WildVoices.png", label="Wild Voices", show_label=False)
        with gr.Column(scale=0.75):
            gr.Markdown("""
            
            # 🌲 **Wild Voices** — *Listening to the More-than-Human World*  
            Welcome to **Wild Voices**, a unique space where you can converse with the more-than-human world. 
            Here, you are invited to ask questions to *rivers*, *trees*, *owls*, *foxes*, and many more. 
            Listen as they respond from their own perspectives—offering the wisdom of the forest, the resilience of the river, and the gentle whispers of the wind.  
            
            🦄 **Generate Myths and Facts:** Challenge common narratives with our *Myth/Fact Generator*, guided by nature’s voice of truth.            
            🎲 **Ask Random Questions:** Get inspired by thought-provoking questions that spark connection with the natural world.  
            🦉 **Discover Hidden Wisdom:** Experience the reflections of *Oak*, *Dragonfly*, *Rain*, and even the humble *Dandelion* as they share their stories. 
            
            ---              
            **Space created and powered by [The H4rmony Project](https://TheH4rmonyproject.org)** — Promoting Sustainable Narratives Through AI. 
            
            _Personae, questions and myth/fact datasets have been generated by [Theophrastus](https://chatgpt.com/g/g-XKAVRvxwc-theophrastus), a H4rmony chat assistant._             
            _Based on an original concept by [Crystal Campbell](https://www.linkedin.com/in/earthly/) for a more-than-human AI Council of Beings._ 
            """)

            
    with gr.Row():
        with gr.Column(scale=0.15): 
            persona1 = gr.Dropdown(personas, label="Choose First Persona", value="Earth")
        with gr.Column(scale=0.15): 
            persona2 = gr.Dropdown(personas, label="Choose Second Persona", value="Crow")          
        user_input = gr.Textbox(label="🌱 Your Question", placeholder="e.g., What do you think of humans?")
    with gr.Row():
        with gr.Column(scale=0.15):   
            pass
        with gr.Column(scale=0.15):   
            pass
        with gr.Column(scale=0.15):      
            random_button = gr.Button("🎲 Generate Random Question") 
        with gr.Column(scale=0.15):  
            ask_button = gr.Button("🌎 Submit Question")
        with gr.Column(scale=0.15):     
            myth_fact_button = gr.Button("🤔 Generate Random Myth/Fact")

    with gr.Row():
        with gr.Column(scale=0.50):
            output1_title = gr.Markdown("### ")
        with gr.Column(scale=0.50):
            output2_title = gr.Markdown("### ")
    with gr.Row():
        output1 = gr.Textbox(label="")
        output2 = gr.Textbox(label="")

    # Button events
    random_button.click(fn=get_random_question, inputs=[], outputs=[user_input])
    
    # Myth/Fact button click event
    myth_fact_button.click(
        fn=get_random_myth_or_fact, 
        inputs=[], 
        outputs=[user_input, persona1, persona2, output1, output2, output1_title, output2_title]
    )

    ask_button.click(
        fn=ask_with_titles, 
        inputs=[persona1, persona2, user_input], 
        outputs=[output1, output2, output1_title, output2_title]
    )

if __name__ == "__main__":
    demo.launch()