Spaces:
Running
Running
Add index.html
Browse files- index.html +192 -0
index.html
ADDED
|
@@ -0,0 +1,192 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<!DOCTYPE html>
|
| 2 |
+
<html>
|
| 3 |
+
<head>
|
| 4 |
+
<meta charset="utf-8" />
|
| 5 |
+
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
| 6 |
+
<title>Story RAG Assistant (Gradio-Lite)</title>
|
| 7 |
+
<script type="module" crossorigin src="https://cdn.jsdelivr.net/npm/@gradio/lite/dist/lite.js"></script>
|
| 8 |
+
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/@gradio/lite/dist/lite.css" />
|
| 9 |
+
<style>
|
| 10 |
+
body {
|
| 11 |
+
background-color: #0f172a;
|
| 12 |
+
color: #f8fafc;
|
| 13 |
+
font-family: 'Inter', system-ui, -apple-system, sans-serif;
|
| 14 |
+
margin: 0;
|
| 15 |
+
padding: 20px;
|
| 16 |
+
}
|
| 17 |
+
</style>
|
| 18 |
+
</head>
|
| 19 |
+
<body>
|
| 20 |
+
<gradio-lite>
|
| 21 |
+
<gradio-code>
|
| 22 |
+
import gradio as gr
|
| 23 |
+
import json
|
| 24 |
+
from pyodide.http import pyfetch
|
| 25 |
+
|
| 26 |
+
async def make_post_request(url, headers, body_dict):
|
| 27 |
+
response = await pyfetch(
|
| 28 |
+
url,
|
| 29 |
+
method="POST",
|
| 30 |
+
headers=headers,
|
| 31 |
+
body=json.dumps(body_dict)
|
| 32 |
+
)
|
| 33 |
+
if response.status != 200:
|
| 34 |
+
err_msg = await response.string()
|
| 35 |
+
raise Exception(f"Request failed with status {response.status}: {err_msg}")
|
| 36 |
+
return await response.json()
|
| 37 |
+
|
| 38 |
+
async def get_pinecone_host(api_key, index_name):
|
| 39 |
+
# Route control plane through corsproxy.io because Pinecone blocks browser CORS
|
| 40 |
+
url = f"https://corsproxy.io/?url=https://api.pinecone.io/indexes/{index_name}"
|
| 41 |
+
response = await pyfetch(
|
| 42 |
+
url,
|
| 43 |
+
method="GET",
|
| 44 |
+
headers={
|
| 45 |
+
"Api-Key": api_key,
|
| 46 |
+
"Accept": "application/json"
|
| 47 |
+
}
|
| 48 |
+
)
|
| 49 |
+
if response.status != 200:
|
| 50 |
+
err_msg = await response.string()
|
| 51 |
+
raise Exception(f"Failed to get Pinecone index info: {err_msg}")
|
| 52 |
+
data = await response.json()
|
| 53 |
+
return data["host"]
|
| 54 |
+
|
| 55 |
+
async def query_pinecone(api_key, host, namespace, text, top_k=4):
|
| 56 |
+
# Route data plane through corsproxy.io
|
| 57 |
+
url = f"https://corsproxy.io/?url=https://{host}/query"
|
| 58 |
+
headers = {
|
| 59 |
+
"Api-Key": api_key,
|
| 60 |
+
"Content-Type": "application/json",
|
| 61 |
+
"Accept": "application/json"
|
| 62 |
+
}
|
| 63 |
+
body = {
|
| 64 |
+
"namespace": namespace,
|
| 65 |
+
"topK": top_k,
|
| 66 |
+
"inputs": {
|
| 67 |
+
"text": text
|
| 68 |
+
},
|
| 69 |
+
"includeMetadata": True
|
| 70 |
+
}
|
| 71 |
+
data = await make_post_request(url, headers, body)
|
| 72 |
+
matches = data.get("matches", [])
|
| 73 |
+
chunks = [m["metadata"].get("chunk_text", "") for m in matches if "metadata" in m]
|
| 74 |
+
return chunks
|
| 75 |
+
|
| 76 |
+
async def call_gemini(api_key, model_name, contents):
|
| 77 |
+
# Gemini API natively supports CORS, so we do not need a proxy for it
|
| 78 |
+
url = f"https://generativelanguage.googleapis.com/v1beta/models/{model_name}:generateContent?key={api_key}"
|
| 79 |
+
headers = {
|
| 80 |
+
"Content-Type": "application/json"
|
| 81 |
+
}
|
| 82 |
+
body = {
|
| 83 |
+
"contents": contents,
|
| 84 |
+
"generationConfig": {
|
| 85 |
+
"temperature": 0.1
|
| 86 |
+
}
|
| 87 |
+
}
|
| 88 |
+
data = await make_post_request(url, headers, body)
|
| 89 |
+
try:
|
| 90 |
+
text = data["candidates"][0]["content"]["parts"][0]["text"]
|
| 91 |
+
return text
|
| 92 |
+
except (KeyError, IndexError):
|
| 93 |
+
raise Exception(f"Failed to parse Gemini response: {json.dumps(data)}")
|
| 94 |
+
|
| 95 |
+
async def respond(message, history, google_key, pinecone_key, pinecone_index, pinecone_ns, google_model):
|
| 96 |
+
if not google_key or not pinecone_key:
|
| 97 |
+
history.append({"role": "user", "content": message})
|
| 98 |
+
history.append({"role": "assistant", "content": "β Please provide both Google and Pinecone API keys in the configuration panel on the left."})
|
| 99 |
+
return "", history
|
| 100 |
+
|
| 101 |
+
try:
|
| 102 |
+
# 1. Fetch Pinecone Host
|
| 103 |
+
host = await get_pinecone_host(pinecone_key, pinecone_index)
|
| 104 |
+
|
| 105 |
+
# 2. Search Pinecone for context
|
| 106 |
+
chunks = await query_pinecone(pinecone_key, host, pinecone_ns, message)
|
| 107 |
+
|
| 108 |
+
# 3. Format prompt
|
| 109 |
+
sources_text = "\n\n".join(f"Source {i+1}:\n{text}" for i, text in enumerate(chunks)) if chunks else "No relevant sources found."
|
| 110 |
+
|
| 111 |
+
system_prompt = (
|
| 112 |
+
"You are an educational assistant answering questions about stories in uploaded documents.\n"
|
| 113 |
+
"Rules:\n"
|
| 114 |
+
"- Use ONLY facts from the retrieved sources. Never invent details.\n"
|
| 115 |
+
"- Give a direct answer in 2-4 sentences. Do not repeat the question or show your reasoning.\n"
|
| 116 |
+
"- If multiple stories appear, answer about the one most relevant to the question.\n"
|
| 117 |
+
"- If the sources do not contain the answer, say you could not find it in the document."
|
| 118 |
+
)
|
| 119 |
+
|
| 120 |
+
augmented_prompt = (
|
| 121 |
+
f"{system_prompt}\n\n"
|
| 122 |
+
f"Retrieved sources (answer using ONLY these):\n{sources_text}\n\n"
|
| 123 |
+
f"Question: {message}"
|
| 124 |
+
)
|
| 125 |
+
|
| 126 |
+
# 4. Construct conversation contents for Gemini
|
| 127 |
+
contents = []
|
| 128 |
+
for turn in history[-8:]:
|
| 129 |
+
role = "user" if turn["role"] == "user" else "model"
|
| 130 |
+
contents.append({
|
| 131 |
+
"role": role,
|
| 132 |
+
"parts": [{"text": turn["content"]}]
|
| 133 |
+
})
|
| 134 |
+
contents.append({
|
| 135 |
+
"role": "user",
|
| 136 |
+
"parts": [{"text": augmented_prompt}]
|
| 137 |
+
})
|
| 138 |
+
|
| 139 |
+
# 5. Call Gemini
|
| 140 |
+
response_text = await call_gemini(google_key, google_model, contents)
|
| 141 |
+
|
| 142 |
+
history.append({"role": "user", "content": message})
|
| 143 |
+
history.append({"role": "assistant", "content": response_text})
|
| 144 |
+
return "", history
|
| 145 |
+
|
| 146 |
+
except Exception as e:
|
| 147 |
+
history.append({"role": "user", "content": message})
|
| 148 |
+
history.append({"role": "assistant", "content": f"β Error: {str(e)}"})
|
| 149 |
+
return "", history
|
| 150 |
+
|
| 151 |
+
# Custom WebAssembly Theme and Styling
|
| 152 |
+
with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="indigo")) as demo:
|
| 153 |
+
gr.Markdown("# π Static RAG Story Assistant (Gradio-Lite)")
|
| 154 |
+
gr.Markdown(
|
| 155 |
+
"This application runs **entirely inside your browser** via WebAssembly (Pyodide). "
|
| 156 |
+
"No backend server is required, and your API keys never leave your browser session (except to call Pinecone and Gemini)."
|
| 157 |
+
)
|
| 158 |
+
|
| 159 |
+
with gr.Row():
|
| 160 |
+
with gr.Column(scale=1):
|
| 161 |
+
gr.Markdown("### π API Configuration")
|
| 162 |
+
google_key_input = gr.Textbox(label="Google API Key", type="password", placeholder="AIzaSy...")
|
| 163 |
+
pinecone_key_input = gr.Textbox(label="Pinecone API Key", type="password", placeholder="pcsk_...")
|
| 164 |
+
pinecone_index_input = gr.Textbox(label="Pinecone Index Name", value="story-llama")
|
| 165 |
+
pinecone_ns_input = gr.Textbox(label="Pinecone Namespace", value="default")
|
| 166 |
+
google_model_input = gr.Textbox(label="Google Model", value="gemini-2.0-flash")
|
| 167 |
+
|
| 168 |
+
with gr.Column(scale=2):
|
| 169 |
+
chatbot = gr.Chatbot(type="messages", height=450)
|
| 170 |
+
msg_input = gr.Textbox(placeholder="Ask a question about the story...", label="Your Question")
|
| 171 |
+
|
| 172 |
+
with gr.Row():
|
| 173 |
+
submit_btn = gr.Button("Send", variant="primary")
|
| 174 |
+
clear_btn = gr.Button("Clear Chat")
|
| 175 |
+
|
| 176 |
+
submit_btn.click(
|
| 177 |
+
respond,
|
| 178 |
+
inputs=[msg_input, chatbot, google_key_input, pinecone_key_input, pinecone_index_input, pinecone_ns_input, google_model_input],
|
| 179 |
+
outputs=[msg_input, chatbot]
|
| 180 |
+
)
|
| 181 |
+
msg_input.submit(
|
| 182 |
+
respond,
|
| 183 |
+
inputs=[msg_input, chatbot, google_key_input, pinecone_key_input, pinecone_index_input, pinecone_ns_input, google_model_input],
|
| 184 |
+
outputs=[msg_input, chatbot]
|
| 185 |
+
)
|
| 186 |
+
clear_btn.click(lambda: [], None, chatbot)
|
| 187 |
+
|
| 188 |
+
demo.launch()
|
| 189 |
+
</gradio-code>
|
| 190 |
+
</gradio-lite>
|
| 191 |
+
</body>
|
| 192 |
+
</html>
|