Spaces:
Sleeping
Sleeping
Upload 9 files
Browse files- .dockerignore +5 -0
- .env +2 -0
- .firebaserc +5 -0
- Dockerfile +22 -0
- README.md +8 -7
- core_logic.py +225 -0
- firebase.json +19 -0
- requirements.txt +7 -0
- server.py +266 -0
.dockerignore
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
__pycache__/
|
| 2 |
+
*.pyc
|
| 3 |
+
.env
|
| 4 |
+
venv/
|
| 5 |
+
.venv/
|
.env
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Environment variables for the AI Browser backend
|
| 2 |
+
OPENROUTER_API_KEY=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
|
.firebaserc
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"projects": {
|
| 3 |
+
"default": "verse-ai-browser"
|
| 4 |
+
}
|
| 5 |
+
}
|
Dockerfile
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM python:3.10-slim
|
| 2 |
+
|
| 3 |
+
WORKDIR /app
|
| 4 |
+
|
| 5 |
+
# Install system dependencies required for ML libraries
|
| 6 |
+
RUN apt-get update && apt-get install -y \
|
| 7 |
+
build-essential \
|
| 8 |
+
&& rm -rf /var/lib/apt/lists/*
|
| 9 |
+
|
| 10 |
+
# Install Python dependencies
|
| 11 |
+
COPY requirements.txt .
|
| 12 |
+
RUN pip install --no-cache-dir -r requirements.txt
|
| 13 |
+
|
| 14 |
+
# Copy backend code
|
| 15 |
+
COPY . .
|
| 16 |
+
|
| 17 |
+
# Expose port (Hugging Face Spaces defaults to 7860)
|
| 18 |
+
ENV PORT=7860
|
| 19 |
+
EXPOSE 7860
|
| 20 |
+
|
| 21 |
+
# Run the FastAPI server
|
| 22 |
+
CMD ["uvicorn", "server:app", "--host", "0.0.0.0", "--port", "7860"]
|
README.md
CHANGED
|
@@ -1,11 +1,12 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
-
colorTo:
|
| 6 |
sdk: docker
|
| 7 |
-
|
| 8 |
-
license: mit
|
| 9 |
---
|
| 10 |
|
| 11 |
-
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: AI Browser Backend
|
| 3 |
+
emoji: 🚀
|
| 4 |
+
colorFrom: blue
|
| 5 |
+
colorTo: purple
|
| 6 |
sdk: docker
|
| 7 |
+
app_port: 7860
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# AI Browser Backend
|
| 11 |
+
|
| 12 |
+
This is the FastAPI backend for the AI Browser, powered by `sentence-transformers` and `torch`.
|
core_logic.py
ADDED
|
@@ -0,0 +1,225 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
# pyrefly: ignore [missing-import]
|
| 3 |
+
import numpy as np
|
| 4 |
+
# pyrefly: ignore [missing-import]
|
| 5 |
+
from bs4 import BeautifulSoup
|
| 6 |
+
# pyrefly: ignore [missing-import]
|
| 7 |
+
from openai import OpenAI
|
| 8 |
+
|
| 9 |
+
_model = None
|
| 10 |
+
|
| 11 |
+
def get_model():
|
| 12 |
+
"""Lazily load the SentenceTransformer model to save memory/startup time."""
|
| 13 |
+
global _model
|
| 14 |
+
if _model is None:
|
| 15 |
+
# pyrefly: ignore [missing-import]
|
| 16 |
+
from sentence_transformers import SentenceTransformer
|
| 17 |
+
_model = SentenceTransformer('all-MiniLM-L6-v2')
|
| 18 |
+
return _model
|
| 19 |
+
|
| 20 |
+
def extract_clean_prose(html_content: str) -> str:
|
| 21 |
+
"""Extract clean prose text from HTML, stripping script, style, and nav tags."""
|
| 22 |
+
if not html_content:
|
| 23 |
+
return ""
|
| 24 |
+
soup = BeautifulSoup(html_content, 'html.parser')
|
| 25 |
+
for tag in soup(["script", "style", "nav"]):
|
| 26 |
+
tag.decompose()
|
| 27 |
+
raw_text = soup.get_text(separator=' ')
|
| 28 |
+
clean_text = ' '.join(raw_text.split())
|
| 29 |
+
return clean_text
|
| 30 |
+
|
| 31 |
+
def chunk_text(text: str, chunk_size: int = 50, overlap: int = 10) -> list[str]:
|
| 32 |
+
"""Chunk text using a sliding window based on word count."""
|
| 33 |
+
words = text.split()
|
| 34 |
+
if not words:
|
| 35 |
+
return []
|
| 36 |
+
chunks = []
|
| 37 |
+
step = chunk_size - overlap
|
| 38 |
+
if step <= 0:
|
| 39 |
+
step = 1 # Avoid infinite loop
|
| 40 |
+
|
| 41 |
+
for i in range(0, len(words), step):
|
| 42 |
+
chunk_words = words[i:i + chunk_size]
|
| 43 |
+
chunk_text = ' '.join(chunk_words)
|
| 44 |
+
chunks.append(chunk_text)
|
| 45 |
+
if i + chunk_size >= len(words):
|
| 46 |
+
break
|
| 47 |
+
|
| 48 |
+
return chunks
|
| 49 |
+
|
| 50 |
+
def embed_texts(texts: list[str]) -> list[list[float]]:
|
| 51 |
+
"""Generate embeddings for a list of texts using SentenceTransformer."""
|
| 52 |
+
if not texts:
|
| 53 |
+
return []
|
| 54 |
+
model = get_model()
|
| 55 |
+
embeddings = model.encode(texts, show_progress_bar=False)
|
| 56 |
+
# Convert numpy array to standard python list of lists
|
| 57 |
+
return embeddings.tolist()
|
| 58 |
+
|
| 59 |
+
def cosine_similarity(v1, v2) -> float:
|
| 60 |
+
"""Compute cosine similarity between two vectors."""
|
| 61 |
+
dot_product = np.dot(v1, v2)
|
| 62 |
+
norm_v1 = np.linalg.norm(v1)
|
| 63 |
+
norm_v2 = np.linalg.norm(v2)
|
| 64 |
+
if norm_v1 == 0 or norm_v2 == 0:
|
| 65 |
+
return 0.0
|
| 66 |
+
return float(dot_product / (norm_v1 * norm_v2))
|
| 67 |
+
|
| 68 |
+
def retrieve_top_chunks(query: str, chunks: list[str], chunk_embeddings: list[list[float]], top_k: int = 2) -> list[dict]:
|
| 69 |
+
"""Retrieve top_k matching chunks for a query using cosine similarity."""
|
| 70 |
+
if not chunks or not chunk_embeddings:
|
| 71 |
+
return []
|
| 72 |
+
|
| 73 |
+
# Generate query embedding
|
| 74 |
+
model = get_model()
|
| 75 |
+
query_emb = model.encode(query, show_progress_bar=False)
|
| 76 |
+
|
| 77 |
+
scores = []
|
| 78 |
+
for idx, emb_list in enumerate(chunk_embeddings):
|
| 79 |
+
emb = np.array(emb_list)
|
| 80 |
+
score = cosine_similarity(query_emb, emb)
|
| 81 |
+
scores.append({
|
| 82 |
+
"score": score,
|
| 83 |
+
"index": idx,
|
| 84 |
+
"text": chunks[idx]
|
| 85 |
+
})
|
| 86 |
+
|
| 87 |
+
# Sort by score descending
|
| 88 |
+
scores.sort(key=lambda x: x["score"], reverse=True)
|
| 89 |
+
return scores[:top_k]
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
# --- LLM Helper ---
|
| 93 |
+
|
| 94 |
+
def _call_llm(system_prompt: str, user_text: str, api_key: str | None = None) -> str:
|
| 95 |
+
"""Internal helper to call OpenRouter LLM with a given system prompt and user text."""
|
| 96 |
+
if not api_key:
|
| 97 |
+
api_key = os.environ.get("OPENROUTER_API_KEY")
|
| 98 |
+
if not api_key:
|
| 99 |
+
raise ValueError(
|
| 100 |
+
"No API key provided. Set the OPENROUTER_API_KEY environment variable "
|
| 101 |
+
"or pass api_key in the request."
|
| 102 |
+
)
|
| 103 |
+
|
| 104 |
+
client = OpenAI(
|
| 105 |
+
base_url="https://openrouter.ai/api/v1",
|
| 106 |
+
api_key=api_key,
|
| 107 |
+
)
|
| 108 |
+
|
| 109 |
+
models_to_try = [
|
| 110 |
+
"openai/gpt-oss-120b:free",
|
| 111 |
+
"google/gemma-4-31b-it:free",
|
| 112 |
+
"nvidia/nemotron-3-super-120b-a12b:free",
|
| 113 |
+
"qwen/qwen3-next-80b-a3b-instruct:free"
|
| 114 |
+
]
|
| 115 |
+
|
| 116 |
+
last_error = None
|
| 117 |
+
for model in models_to_try:
|
| 118 |
+
try:
|
| 119 |
+
response = client.chat.completions.create(
|
| 120 |
+
model=model,
|
| 121 |
+
messages=[
|
| 122 |
+
{"role": "system", "content": system_prompt},
|
| 123 |
+
{"role": "user", "content": user_text}
|
| 124 |
+
]
|
| 125 |
+
)
|
| 126 |
+
return response.choices[0].message.content
|
| 127 |
+
except Exception as e:
|
| 128 |
+
last_error = e
|
| 129 |
+
continue
|
| 130 |
+
|
| 131 |
+
# If all models fail, return a polite fallback message instead of crashing with a 500 error
|
| 132 |
+
error_str = str(last_error)
|
| 133 |
+
if "429" in error_str or "rate-limited" in error_str.lower():
|
| 134 |
+
return "I am currently overloaded with requests (Rate Limited by the provider). Please try clicking the action button again in a few seconds!"
|
| 135 |
+
else:
|
| 136 |
+
return "I encountered an error connecting to the AI provider. Please try again later."
|
| 137 |
+
|
| 138 |
+
|
| 139 |
+
def query_llm(query: str, context: str, api_key: str | None = None, title: str = "", url: str = "", selected_text: str = "") -> str:
|
| 140 |
+
"""Use OpenRouter LLM to answer a query based on the provided context and metadata."""
|
| 141 |
+
system_prompt = (
|
| 142 |
+
"You are an AI assistant built into the 'Verse' browser. Answer accurately using the provided webpage context. "
|
| 143 |
+
"Be extremely concise and direct. Output NO introductory or conversational filler text."
|
| 144 |
+
)
|
| 145 |
+
|
| 146 |
+
metadata_lines = []
|
| 147 |
+
if url:
|
| 148 |
+
metadata_lines.append(f"Page URL: {url}")
|
| 149 |
+
if title:
|
| 150 |
+
metadata_lines.append(f"Page Title: {title}")
|
| 151 |
+
if selected_text:
|
| 152 |
+
metadata_lines.append(f"User's Highlighted Text: {selected_text}")
|
| 153 |
+
|
| 154 |
+
metadata_block = "\n".join(metadata_lines)
|
| 155 |
+
if metadata_block:
|
| 156 |
+
metadata_block = f"--- Page Metadata ---\n{metadata_block}\n---------------------\n\n"
|
| 157 |
+
|
| 158 |
+
user_text = f"{metadata_block}Context Snippets:\n{context}\n\nUser Query: {query}"
|
| 159 |
+
return _call_llm(system_prompt, user_text, api_key)
|
| 160 |
+
|
| 161 |
+
|
| 162 |
+
# --- AI Action Functions ---
|
| 163 |
+
|
| 164 |
+
def summarize_text(text: str, api_key: str | None = None) -> str:
|
| 165 |
+
"""Summarize the given text concisely."""
|
| 166 |
+
system_prompt = (
|
| 167 |
+
"Summarize concisely using short bullet points. Output NO filler or conversational text."
|
| 168 |
+
)
|
| 169 |
+
return _call_llm(system_prompt, text, api_key)
|
| 170 |
+
|
| 171 |
+
def explain_text(text: str, api_key: str | None = None) -> str:
|
| 172 |
+
"""Explain the given text in simple terms."""
|
| 173 |
+
system_prompt = (
|
| 174 |
+
"Explain simply for a beginner. Be extremely concise. Output NO filler text."
|
| 175 |
+
)
|
| 176 |
+
return _call_llm(system_prompt, text, api_key)
|
| 177 |
+
|
| 178 |
+
def translate_text(text: str, target_language: str, api_key: str | None = None) -> str:
|
| 179 |
+
"""Translate text to the target language."""
|
| 180 |
+
system_prompt = (
|
| 181 |
+
f"Translate the text to {target_language}. Output ONLY the raw translation with NO introductory text."
|
| 182 |
+
)
|
| 183 |
+
return _call_llm(system_prompt, text, api_key)
|
| 184 |
+
|
| 185 |
+
def extract_data(text: str, data_type: str, api_key: str | None = None) -> str:
|
| 186 |
+
"""Extract specific data (emails, phones, links) from text."""
|
| 187 |
+
system_prompt = (
|
| 188 |
+
f"Extract all {data_type}. Output ONLY a clean list, one item per line. No filler."
|
| 189 |
+
)
|
| 190 |
+
return _call_llm(system_prompt, text, api_key)
|
| 191 |
+
|
| 192 |
+
def generate_post(text: str, platform: str, api_key: str | None = None) -> str:
|
| 193 |
+
"""Generate a social media post from the text."""
|
| 194 |
+
system_prompt = (
|
| 195 |
+
f"Create a short, engaging {platform} post based on the text. Include hashtags. Output ONLY the post content."
|
| 196 |
+
)
|
| 197 |
+
return _call_llm(system_prompt, text, api_key)
|
| 198 |
+
|
| 199 |
+
def rewrite_text(text: str, style: str, api_key: str | None = None) -> str:
|
| 200 |
+
"""Rewrite text in a specified style."""
|
| 201 |
+
system_prompt = (
|
| 202 |
+
f"Rewrite in a {style} style. Output ONLY the rewritten text with NO filler."
|
| 203 |
+
)
|
| 204 |
+
return _call_llm(system_prompt, text, api_key)
|
| 205 |
+
|
| 206 |
+
def list_key_points(text: str, api_key: str | None = None) -> str:
|
| 207 |
+
"""Extract key points from the text."""
|
| 208 |
+
system_prompt = (
|
| 209 |
+
"Extract key points as a numbered list. Be extremely brief. Output NO filler text."
|
| 210 |
+
)
|
| 211 |
+
return _call_llm(system_prompt, text, api_key)
|
| 212 |
+
|
| 213 |
+
def generate_faqs(text: str, api_key: str | None = None) -> str:
|
| 214 |
+
"""Generate FAQs from the text."""
|
| 215 |
+
system_prompt = (
|
| 216 |
+
"Generate 3-5 FAQs based on the text. Format as 'Q: ...' 'A: ...'. Keep answers strictly concise. Output NO filler."
|
| 217 |
+
)
|
| 218 |
+
return _call_llm(system_prompt, text, api_key)
|
| 219 |
+
|
| 220 |
+
def create_study_notes(text: str, api_key: str | None = None) -> str:
|
| 221 |
+
"""Create study notes from the text."""
|
| 222 |
+
system_prompt = (
|
| 223 |
+
"Create brief study notes with headings and bullet points. Be extremely concise. Output NO filler text."
|
| 224 |
+
)
|
| 225 |
+
return _call_llm(system_prompt, text, api_key)
|
firebase.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hosting": {
|
| 3 |
+
"public": "public",
|
| 4 |
+
"ignore": [
|
| 5 |
+
"firebase.json",
|
| 6 |
+
"**/.*",
|
| 7 |
+
"**/node_modules/**"
|
| 8 |
+
],
|
| 9 |
+
"rewrites": [
|
| 10 |
+
{
|
| 11 |
+
"source": "**",
|
| 12 |
+
"run": {
|
| 13 |
+
"serviceId": "ai-browser-backend",
|
| 14 |
+
"region": "us-central1"
|
| 15 |
+
}
|
| 16 |
+
}
|
| 17 |
+
]
|
| 18 |
+
}
|
| 19 |
+
}
|
requirements.txt
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
fastapi>=0.100.0
|
| 2 |
+
uvicorn>=0.22.0
|
| 3 |
+
python-dotenv>=1.0.0
|
| 4 |
+
beautifulsoup4>=4.12.0
|
| 5 |
+
sentence-transformers>=2.2.0
|
| 6 |
+
numpy>=1.20.0
|
| 7 |
+
openai>=1.0.0
|
server.py
ADDED
|
@@ -0,0 +1,266 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# pyrefly: ignore [missing-import]
|
| 2 |
+
from dotenv import load_dotenv
|
| 3 |
+
load_dotenv()
|
| 4 |
+
|
| 5 |
+
import os
|
| 6 |
+
# pyrefly: ignore [missing-import]
|
| 7 |
+
from fastapi import FastAPI, HTTPException
|
| 8 |
+
# pyrefly: ignore [missing-import]
|
| 9 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 10 |
+
# pyrefly: ignore [missing-import]
|
| 11 |
+
from pydantic import BaseModel, Field
|
| 12 |
+
|
| 13 |
+
# Import core RAG logic functions
|
| 14 |
+
import core_logic
|
| 15 |
+
|
| 16 |
+
app = FastAPI(
|
| 17 |
+
title="AI Browser RAG Backend",
|
| 18 |
+
description="Backend API for text parsing, embedding, retrieval, LLM querying, and AI actions",
|
| 19 |
+
version="2.0.0"
|
| 20 |
+
)
|
| 21 |
+
|
| 22 |
+
# Allow CORS for all origins since it will be deployed
|
| 23 |
+
app.add_middleware(
|
| 24 |
+
CORSMiddleware,
|
| 25 |
+
allow_origins=["*"],
|
| 26 |
+
allow_credentials=True,
|
| 27 |
+
allow_methods=["*"],
|
| 28 |
+
allow_headers=["*"],
|
| 29 |
+
)
|
| 30 |
+
|
| 31 |
+
# --- Pydantic Schemas ---
|
| 32 |
+
|
| 33 |
+
class ProcessHtmlRequest(BaseModel):
|
| 34 |
+
html: str
|
| 35 |
+
chunk_size: int = Field(default=50, ge=1)
|
| 36 |
+
overlap: int = Field(default=10, ge=0)
|
| 37 |
+
|
| 38 |
+
class ProcessHtmlResponse(BaseModel):
|
| 39 |
+
chunks: list[str]
|
| 40 |
+
|
| 41 |
+
class EmbedRequest(BaseModel):
|
| 42 |
+
texts: list[str]
|
| 43 |
+
|
| 44 |
+
class EmbedResponse(BaseModel):
|
| 45 |
+
embeddings: list[list[float]]
|
| 46 |
+
|
| 47 |
+
class RetrieveRequest(BaseModel):
|
| 48 |
+
query: str
|
| 49 |
+
chunks: list[str]
|
| 50 |
+
embeddings: list[list[float]]
|
| 51 |
+
top_k: int = Field(default=2, ge=1)
|
| 52 |
+
|
| 53 |
+
class RetrievalResultItem(BaseModel):
|
| 54 |
+
score: float
|
| 55 |
+
index: int
|
| 56 |
+
text: str
|
| 57 |
+
|
| 58 |
+
class RetrieveResponse(BaseModel):
|
| 59 |
+
results: list[RetrievalResultItem]
|
| 60 |
+
|
| 61 |
+
class AnswerRequest(BaseModel):
|
| 62 |
+
query: str
|
| 63 |
+
context: str
|
| 64 |
+
api_key: str | None = None
|
| 65 |
+
|
| 66 |
+
class AnswerResponse(BaseModel):
|
| 67 |
+
answer: str
|
| 68 |
+
|
| 69 |
+
class UnifiedRagRequest(BaseModel):
|
| 70 |
+
html: str
|
| 71 |
+
query: str
|
| 72 |
+
chunk_size: int = Field(default=50, ge=1)
|
| 73 |
+
overlap: int = Field(default=10, ge=0)
|
| 74 |
+
top_k: int = Field(default=2, ge=1)
|
| 75 |
+
title: str = ""
|
| 76 |
+
url: str = ""
|
| 77 |
+
selected_text: str = ""
|
| 78 |
+
api_key: str | None = None
|
| 79 |
+
|
| 80 |
+
class UnifiedRagResponse(BaseModel):
|
| 81 |
+
answer: str
|
| 82 |
+
retrieved_chunks: list[RetrievalResultItem]
|
| 83 |
+
|
| 84 |
+
# --- AI Action Schemas ---
|
| 85 |
+
|
| 86 |
+
class TextRequest(BaseModel):
|
| 87 |
+
text: str
|
| 88 |
+
api_key: str | None = None
|
| 89 |
+
|
| 90 |
+
class TranslateRequest(BaseModel):
|
| 91 |
+
text: str
|
| 92 |
+
target_language: str
|
| 93 |
+
api_key: str | None = None
|
| 94 |
+
|
| 95 |
+
class ExtractRequest(BaseModel):
|
| 96 |
+
text: str
|
| 97 |
+
data_type: str
|
| 98 |
+
api_key: str | None = None
|
| 99 |
+
|
| 100 |
+
class GeneratePostRequest(BaseModel):
|
| 101 |
+
text: str
|
| 102 |
+
platform: str
|
| 103 |
+
api_key: str | None = None
|
| 104 |
+
|
| 105 |
+
class RewriteRequest(BaseModel):
|
| 106 |
+
text: str
|
| 107 |
+
style: str
|
| 108 |
+
api_key: str | None = None
|
| 109 |
+
|
| 110 |
+
class ActionResponse(BaseModel):
|
| 111 |
+
result: str
|
| 112 |
+
|
| 113 |
+
# --- Route Handlers ---
|
| 114 |
+
|
| 115 |
+
@app.get("/")
|
| 116 |
+
def read_root():
|
| 117 |
+
return {"status": "ok", "message": "AI Browser RAG Backend API is running.", "version": "2.0.0"}
|
| 118 |
+
|
| 119 |
+
@app.post("/process-html", response_model=ProcessHtmlResponse)
|
| 120 |
+
def process_html(payload: ProcessHtmlRequest):
|
| 121 |
+
try:
|
| 122 |
+
clean_text = core_logic.extract_clean_prose(payload.html)
|
| 123 |
+
chunks = core_logic.chunk_text(clean_text, payload.chunk_size, payload.overlap)
|
| 124 |
+
return ProcessHtmlResponse(chunks=chunks)
|
| 125 |
+
except Exception as e:
|
| 126 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 127 |
+
|
| 128 |
+
@app.post("/embed", response_model=EmbedResponse)
|
| 129 |
+
def embed_chunks(payload: EmbedRequest):
|
| 130 |
+
try:
|
| 131 |
+
embeddings = core_logic.embed_texts(payload.texts)
|
| 132 |
+
return EmbedResponse(embeddings=embeddings)
|
| 133 |
+
except Exception as e:
|
| 134 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 135 |
+
|
| 136 |
+
@app.post("/retrieve", response_model=RetrieveResponse)
|
| 137 |
+
def retrieve(payload: RetrieveRequest):
|
| 138 |
+
try:
|
| 139 |
+
if len(payload.chunks) != len(payload.embeddings):
|
| 140 |
+
raise HTTPException(status_code=400, detail="The number of chunks must match the number of embeddings.")
|
| 141 |
+
results = core_logic.retrieve_top_chunks(payload.query, payload.chunks, payload.embeddings, payload.top_k)
|
| 142 |
+
return RetrieveResponse(results=results)
|
| 143 |
+
except HTTPException as he:
|
| 144 |
+
raise he
|
| 145 |
+
except Exception as e:
|
| 146 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 147 |
+
|
| 148 |
+
@app.post("/answer", response_model=AnswerResponse)
|
| 149 |
+
def get_answer(payload: AnswerRequest):
|
| 150 |
+
try:
|
| 151 |
+
answer = core_logic.query_llm(payload.query, payload.context, payload.api_key)
|
| 152 |
+
return AnswerResponse(answer=answer)
|
| 153 |
+
except Exception as e:
|
| 154 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 155 |
+
|
| 156 |
+
@app.post("/rag", response_model=UnifiedRagResponse)
|
| 157 |
+
def run_unified_rag(payload: UnifiedRagRequest):
|
| 158 |
+
try:
|
| 159 |
+
# Step 1: Clean and chunk HTML
|
| 160 |
+
clean_text = core_logic.extract_clean_prose(payload.html)
|
| 161 |
+
chunks = core_logic.chunk_text(clean_text, payload.chunk_size, payload.overlap)
|
| 162 |
+
|
| 163 |
+
if not chunks:
|
| 164 |
+
return UnifiedRagResponse(answer="No text chunks could be extracted from the HTML content.", retrieved_chunks=[])
|
| 165 |
+
|
| 166 |
+
# Step 2: Embed chunks
|
| 167 |
+
embeddings = core_logic.embed_texts(chunks)
|
| 168 |
+
|
| 169 |
+
# Step 3: Retrieve top chunks
|
| 170 |
+
top_chunks = core_logic.retrieve_top_chunks(payload.query, chunks, embeddings, payload.top_k)
|
| 171 |
+
|
| 172 |
+
# Step 4: Format context and query LLM
|
| 173 |
+
context = "\n\n".join([f"[Source Chunk {item['index'] + 1}]: {item['text']}" for item in top_chunks])
|
| 174 |
+
answer = core_logic.query_llm(
|
| 175 |
+
query=payload.query,
|
| 176 |
+
context=context,
|
| 177 |
+
title=payload.title,
|
| 178 |
+
url=payload.url,
|
| 179 |
+
selected_text=payload.selected_text,
|
| 180 |
+
api_key=payload.api_key
|
| 181 |
+
)
|
| 182 |
+
|
| 183 |
+
return UnifiedRagResponse(answer=answer, retrieved_chunks=top_chunks)
|
| 184 |
+
except Exception as e:
|
| 185 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 186 |
+
|
| 187 |
+
# --- AI Action Endpoints ---
|
| 188 |
+
|
| 189 |
+
@app.post("/summarize", response_model=ActionResponse)
|
| 190 |
+
def summarize(payload: TextRequest):
|
| 191 |
+
try:
|
| 192 |
+
result = core_logic.summarize_text(payload.text, payload.api_key)
|
| 193 |
+
return ActionResponse(result=result)
|
| 194 |
+
except Exception as e:
|
| 195 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 196 |
+
|
| 197 |
+
@app.post("/explain", response_model=ActionResponse)
|
| 198 |
+
def explain(payload: TextRequest):
|
| 199 |
+
try:
|
| 200 |
+
result = core_logic.explain_text(payload.text, payload.api_key)
|
| 201 |
+
return ActionResponse(result=result)
|
| 202 |
+
except Exception as e:
|
| 203 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 204 |
+
|
| 205 |
+
@app.post("/translate", response_model=ActionResponse)
|
| 206 |
+
def translate(payload: TranslateRequest):
|
| 207 |
+
try:
|
| 208 |
+
result = core_logic.translate_text(payload.text, payload.target_language, payload.api_key)
|
| 209 |
+
return ActionResponse(result=result)
|
| 210 |
+
except Exception as e:
|
| 211 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 212 |
+
|
| 213 |
+
@app.post("/extract", response_model=ActionResponse)
|
| 214 |
+
def extract(payload: ExtractRequest):
|
| 215 |
+
try:
|
| 216 |
+
result = core_logic.extract_data(payload.text, payload.data_type, payload.api_key)
|
| 217 |
+
return ActionResponse(result=result)
|
| 218 |
+
except Exception as e:
|
| 219 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 220 |
+
|
| 221 |
+
@app.post("/generate-post", response_model=ActionResponse)
|
| 222 |
+
def generate_post(payload: GeneratePostRequest):
|
| 223 |
+
try:
|
| 224 |
+
result = core_logic.generate_post(payload.text, payload.platform, payload.api_key)
|
| 225 |
+
return ActionResponse(result=result)
|
| 226 |
+
except Exception as e:
|
| 227 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 228 |
+
|
| 229 |
+
@app.post("/rewrite", response_model=ActionResponse)
|
| 230 |
+
def rewrite(payload: RewriteRequest):
|
| 231 |
+
try:
|
| 232 |
+
result = core_logic.rewrite_text(payload.text, payload.style, payload.api_key)
|
| 233 |
+
return ActionResponse(result=result)
|
| 234 |
+
except Exception as e:
|
| 235 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 236 |
+
|
| 237 |
+
@app.post("/key-points", response_model=ActionResponse)
|
| 238 |
+
def key_points(payload: TextRequest):
|
| 239 |
+
try:
|
| 240 |
+
result = core_logic.list_key_points(payload.text, payload.api_key)
|
| 241 |
+
return ActionResponse(result=result)
|
| 242 |
+
except Exception as e:
|
| 243 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 244 |
+
|
| 245 |
+
@app.post("/faqs", response_model=ActionResponse)
|
| 246 |
+
def faqs(payload: TextRequest):
|
| 247 |
+
try:
|
| 248 |
+
result = core_logic.generate_faqs(payload.text, payload.api_key)
|
| 249 |
+
return ActionResponse(result=result)
|
| 250 |
+
except Exception as e:
|
| 251 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 252 |
+
|
| 253 |
+
@app.post("/study-notes", response_model=ActionResponse)
|
| 254 |
+
def study_notes(payload: TextRequest):
|
| 255 |
+
try:
|
| 256 |
+
result = core_logic.create_study_notes(payload.text, payload.api_key)
|
| 257 |
+
return ActionResponse(result=result)
|
| 258 |
+
except Exception as e:
|
| 259 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 260 |
+
|
| 261 |
+
if __name__ == "__main__":
|
| 262 |
+
# pyrefly: if __name__ == "__main__":
|
| 263 |
+
import uvicorn
|
| 264 |
+
import os
|
| 265 |
+
port = int(os.environ.get("PORT", 8000))
|
| 266 |
+
uvicorn.run(app, host="0.0.0.0", port=port)
|