#!/usr/bin/env bash
set -euo pipefail
# --------------------
# Minimal zero-edit installer + runner for Cognito
# --------------------
# Usage:
# ./setup_and_run.sh
# Optional env overrides:
# HF_API_TOKEN - your Hugging Face inference API token (if omitted, server runs offline fallback)
# HF_MODEL_ID - model id for HF inference (default: gpt2)
# COGNITO_API_KEY - API key used by the widget to call /chat (default provided)
# ALLOW_FETCH - "true" to allow server-side fetches (default false)
# FETCH_BEARER_TOKEN - optional bearer token used by server to fetch authorized pages (if you provide)
#
# Note: Do NOT use this to bypass site protections. Only fetch pages you own/are authorized to access.
ROOT_DIR="$(pwd)"
VENV_DIR="$ROOT_DIR/.venv"
echo "==> Preparing virtualenv..."
if ! command -v python3 >/dev/null 2>&1; then
echo "Python3 is required. Install Python 3.10+ and retry."
exit 1
fi
python3 -m venv "$VENV_DIR"
# shellcheck disable=SC1091
source "$VENV_DIR/bin/activate"
pip install --upgrade pip
cat > requirements.txt <<'PYREQ'
fastapi==0.95.2
uvicorn[standard]==0.21.1
httpx==0.24.1
python-dotenv==1.0.0
python-multipart==0.0.6
sqlalchemy==2.0.19
pydantic==1.10.12
PYREQ
echo "==> Installing Python dependencies (lightweight)..."
pip install -r requirements.txt
# Write app.py
cat > app.py <<'PYAPP'
import os
import json
import sqlite3
import urllib.parse
from typing import Optional, List, Dict
from fastapi import FastAPI, Request, HTTPException, UploadFile, File, Form
from pydantic import BaseModel
import httpx
from starlette.responses import PlainTextResponse, HTMLResponse, Response, FileResponse
# Config (can be overridden by environment variables)
HF_API_TOKEN = os.getenv("HF_API_TOKEN", "")
HF_MODEL_ID = os.getenv("HF_MODEL_ID", "gpt2")
COGNITO_API_KEY = os.getenv("COGNITO_API_KEY", "cognito_default_key_please_change")
ALLOW_FETCH = os.getenv("ALLOW_FETCH", "false").lower() == "true"
FETCH_BEARER_TOKEN = os.getenv("FETCH_BEARER_TOKEN", "")
MODERATION_BLOCKLIST = [s.strip().lower() for s in os.getenv("MODERATION_BLOCKLIST", "hack,steal,illegal").split(",") if s.strip()]
DB_PATH = os.path.join(os.getcwd(), "cognito_sessions.db")
UPLOADS_DIR = os.path.join(os.getcwd(), "uploads")
os.makedirs(UPLOADS_DIR, exist_ok=True)
app = FastAPI(title="Cognito (Minimal)", docs_url="/docs")
# --- DB helpers (SQLite simple storage) ---
def init_db():
conn = sqlite3.connect(DB_PATH)
cur = conn.cursor()
cur.execute("""
CREATE TABLE IF NOT EXISTS sessions (
session_id TEXT PRIMARY KEY,
messages TEXT
)
""")
conn.commit()
conn.close()
def get_session_messages(session_id: str):
conn = sqlite3.connect(DB_PATH)
cur = conn.cursor()
cur.execute("SELECT messages FROM sessions WHERE session_id = ?", (session_id,))
row = cur.fetchone()
conn.close()
if row:
return json.loads(row[0])
return []
def save_session_messages(session_id: str, messages):
conn = sqlite3.connect(DB_PATH)
cur = conn.cursor()
cur.execute("INSERT OR REPLACE INTO sessions (session_id, messages) VALUES (?, ?)", (session_id, json.dumps(messages)))
conn.commit()
conn.close()
init_db()
# --- Request models ---
class ChatRequest(BaseModel):
session_id: Optional[str] = None
messages: List[Dict] # [{"role":"user","content":"..."}]
fetch_url: Optional[str] = None
# --- Moderation ---
def moderate_messages(messages):
txt = " ".join(m.get("content","") for m in messages).lower()
for bad in MODERATION_BLOCKLIST:
if bad and bad in txt:
return False, f"Message blocked by moderation: found banned word '{bad}'"
return True, ""
# --- Helper: simple prompt builder ---
def build_prompt(messages, uploaded_texts=None):
prompt = ""
if uploaded_texts:
prompt += "Context documents:\n"
for i, t in enumerate(uploaded_texts):
prompt += f"[DOC {i+1}]\n{t}\n\n"
prompt += "---\n"
for m in messages:
role = m.get("role","user")
content = m.get("content","")
if role == "system":
prompt += f"[SYSTEM] {content}\n"
elif role == "user":
prompt += f"User: {content}\n"
else:
prompt += f"Assistant: {content}\n"
prompt += "\nAssistant:"
return prompt
# --- File uploads endpoint ---
@app.post("/upload")
async def upload_file(session_id: Optional[str] = Form(None), file: UploadFile = File(...)):
# Stores plain text/markdown files into uploads/, returns file id
content = await file.read()
fname = file.filename
safe_name = fname.replace("..","_").replace("/","_")
path = os.path.join(UPLOADS_DIR, safe_name)
with open(path, "wb") as f:
f.write(content)
return {"ok": True, "path": path}
# --- Serve widget JS ---
@app.get("/cognito-widget.js")
def widget_js():
# This simple widget matches what's described in README and expects window.COGNITO_API_BASE & API key set on host page.
js = """
(function () {
if (window.CognitoWidgetInitialized) return;
window.CognitoWidgetInitialized = true;
const CONFIG = { apiBase: window.COGNITO_API_BASE || "http://localhost:8000", apiKey: window.COGNITO_API_KEY || null, title: "Cognito" };
const css = `.cognito-chat-btn { position: fixed; right: 20px; bottom: 20px; z-index: 100000; background: #0b5fff; color: #fff; border-radius: 28px; padding: 12px 16px; cursor: pointer; } .cognito-panel { position: fixed; right: 20px; bottom: 80px; width: 360px; max-width: 90vw; z-index: 100000; border-radius: 12px; overflow: hidden; background: #fff; display:none; flex-direction:column } .cognito-header{ padding:12px; background:#0b5fff; color:#fff } .cognito-body{ padding:12px; height:300px; overflow:auto; background:#f7f7fb } .cognito-input-row{ display:flex; padding:8px; border-top:1px solid #eee } .cognito-input{ flex:1; padding:8px 10px; border-radius:8px; border:1px solid #ddd } .cognito-send{ margin-left:8px; padding:8px 12px; border-radius:8px; border:none; background:#0b5fff; color:#fff } .cognito-msg{ margin-bottom:8px; padding:8px 10px; border-radius:8px; max-width:85% } .cognito-msg.user{ background:#0b5fff; color:#fff; margin-left:auto } .cognito-msg.bot{ background:#fff; color:#111; border:1px solid #eee }`;
const styleEl=document.createElement("style"); styleEl.innerHTML=css; document.head.appendChild(styleEl);
const btn=document.createElement("button"); btn.className='cognito-chat-btn'; btn.innerText=CONFIG.title; document.body.appendChild(btn);
const panel=document.createElement("div"); panel.className='cognito-panel'; panel.innerHTML=''; document.body.appendChild(panel);
btn.addEventListener('click', ()=>{ panel.style.display = panel.style.display === 'flex' ? 'none' : 'flex'; panel.style.flexDirection='column'; });
const bodyEl=panel.querySelector('.cognito-body'); const inputEl=panel.querySelector('.cognito-input'); const sendBtn=panel.querySelector('.cognito-send');
function appendMessage(text, who){ const m=document.createElement('div'); m.className='cognito-msg '+(who==='user'?'user':'bot'); m.innerText=text; bodyEl.appendChild(m); bodyEl.scrollTop = bodyEl.scrollHeight; }
async function sendMessage(text){ appendMessage(text,'user'); inputEl.value=''; const payload={ messages: [{role:'user', content:text}], session_id: null }; const headers={'Content-Type':'application/json'}; if(CONFIG.apiKey) headers['x-api-key']=CONFIG.apiKey; try{ const res = await fetch(CONFIG.apiBase + '/chat',{ method:'POST', headers, body: JSON.stringify(payload) }); if(!res.ok){ appendMessage('Error: '+res.status+' '+await res.text(),'bot'); return; } const data = await res.json(); appendMessage(data.reply || JSON.stringify(data), 'bot'); } catch(err){ appendMessage('Network error: '+err.message,'bot'); } }
sendBtn.addEventListener('click', ()=>{ const v=inputEl.value.trim(); if(!v) return; sendMessage(v); });
inputEl.addEventListener('keydown', (e)=>{ if(e.key==='Enter'){ e.preventDefault(); sendBtn.click(); } });
window.CognitoWidget = { send: sendMessage, open: () => { panel.style.display='flex'; }, close: () => { panel.style.display='none'; } };
})();
"""
return Response(content=js, media_type="application/javascript")
# --- Status root ---
@app.get("/")
def index():
content = f"Cognito (minimal)
Use the widget by embedding <script src=\"/cognito-widget.js\"></script> and set window.COGNITO_API_BASE & window.COGNITO_API_KEY.
HF_API_TOKEN set: {bool(HF_API_TOKEN)}
"
return HTMLResponse(content=content)
# --- Helper: server-side fetch (ONLY if ALLOW_FETCH enabled) ---
async def fetch_url_with_bearer(url: str):
if not ALLOW_FETCH:
raise HTTPException(status_code=403, detail="Server is not configured to fetch external pages.")
parsed = urllib.parse.urlparse(url)
if parsed.scheme not in ("http","https"):
raise HTTPException(status_code=400, detail="Invalid URL scheme.")
headers = {"User-Agent": "CognitoFetcher/1.0"}
if FETCH_BEARER_TOKEN:
headers["Authorization"] = f"Bearer {FETCH_BEARER_TOKEN}"
async with httpx.AsyncClient(timeout=20.0) as client:
r = await client.get(url, headers=headers)
if r.status_code >= 400:
raise HTTPException(status_code=502, detail=f"Upstream fetch failed: {r.status_code}")
# return text limited to first N chars
return r.text[:20000]
# --- HF inference call (if token provided) ---
def call_hf_inference(prompt: str):
if not HF_API_TOKEN:
# fallback: simple deterministic reply
return "Cognito (offline mode): I received your message. Provide HF_API_TOKEN to enable real model responses."
url = f"https://api-inference.huggingface.co/models/{HF_MODEL_ID}"
headers = {"Authorization": f"Bearer {HF_API_TOKEN}", "Content-Type": "application/json"}
payload = {"inputs": prompt, "options": {"wait_for_model": True}}
try:
with httpx.Client(timeout=60.0) as client:
r = client.post(url, headers=headers, json=payload)
r.raise_for_status()
out = r.json()
# attempt to extract text
if isinstance(out, list) and len(out) > 0 and isinstance(out[0], dict):
return out[0].get("generated_text") or str(out)
if isinstance(out, dict):
return out.get("generated_text") or str(out)
return str(out)
except Exception as e:
return f"Cognito: error calling HF inference: {str(e)}"
# --- /chat endpoint ---
@app.post("/chat")
async def chat(req: ChatRequest, request: Request):
# API key check
header_key = request.headers.get("x-api-key", "")
if header_key != COGNITO_API_KEY:
raise HTTPException(status_code=401, detail="Invalid API key.")
# moderation
ok, reason = moderate_messages(req.messages)
if not ok:
raise HTTPException(status_code=400, detail=reason)
uploaded_texts = []
# If session has uploaded files, read small files from uploads/
if req.session_id:
# naive: read any files that start with session_id in uploads folder
for fname in os.listdir(UPLOADS_DIR):
if fname.startswith(req.session_id):
path = os.path.join(UPLOADS_DIR, fname)
try:
with open(path, "r", encoding="utf-8", errors="replace") as f:
uploaded_texts.append(f.read()[:5000])
except Exception:
pass
# optional fetch
fetched = ""
if req.fetch_url:
fetched = await fetch_url_with_bearer(req.fetch_url)
prompt = build_prompt(req.messages, uploaded_texts + ([fetched] if fetched else []))
reply = call_hf_inference(prompt)
# save to session
sess_id = req.session_id or "anon"
prev = get_session_messages(sess_id)
prev.append({"role":"user","content": req.messages[-1].get("content","") if req.messages else ""})
prev.append({"role":"assistant","content": reply})
# cap history size
if len(prev) > 50:
prev = prev[-50:]
save_session_messages(sess_id, prev)
return {"reply": reply}
# --- simple health route ---
@app.get("/health")
def health():
return {"status": "ok", "hf_token": bool(HF_API_TOKEN), "allow_fetch": ALLOW_FETCH}
PYAPP
# Write cognito-widget.js (also for convenience if user wants standalone file)
cat > cognito-widget.js <<'WIDGET'
/* This file is also available at /cognito-widget.js when server is running.
The hosted widget expects window.COGNITO_API_BASE and window.COGNITO_API_KEY on the host page. */
(function () {
if (window.CognitoWidgetInitialized) return;
window.CognitoWidgetInitialized = true;
const CONFIG = { apiBase: window.COGNITO_API_BASE || "http://localhost:8000", apiKey: window.COGNITO_API_KEY || null, title: "Cognito" };
const css = `.cognito-chat-btn { position: fixed; right: 20px; bottom: 20px; z-index: 100000; background: #0b5fff; color: #fff; border-radius: 28px; padding: 12px 16px; cursor: pointer; } .cognito-panel { position: fixed; right: 20px; bottom: 80px; width: 360px; max-width: 90vw; z-index: 100000; border-radius: 12px; overflow: hidden; background: #fff; display:none; flex-direction:column } .cognito-header{ padding:12px; background:#0b5fff; color:#fff } .cognito-body{ padding:12px; height:300px; overflow:auto; background:#f7f7fb } .cognito-input-row{ display:flex; padding:8px; border-top:1px solid #eee } .cognito-input{ flex:1; padding:8px 10px; border-radius:8px; border:1px solid #ddd } .cognito-send{ margin-left:8px; padding:8px 12px; border-radius:8px; border:none; background:#0b5fff; color:#fff } .cognito-msg{ margin-bottom:8px; padding:8px 10px; border-radius:8px; max-width:85% } .cognito-msg.user{ background:#0b5fff; color:#fff; margin-left:auto } .cognito-msg.bot{ background:#fff; color:#111; border:1px solid #eee }`;
const styleEl=document.createElement("style"); styleEl.innerHTML=css; document.head.appendChild(styleEl);
const btn=document.createElement("button"); btn.className='cognito-chat-btn'; btn.innerText=CONFIG.title; document.body.appendChild(btn);
const panel=document.createElement("div"); panel.className='cognito-panel'; panel.innerHTML=''; document.body.appendChild(panel);
btn.addEventListener('click', ()=>{ panel.style.display = panel.style.display === 'flex' ? 'none' : 'flex'; panel.style.flexDirection='column'; });
const bodyEl=panel.querySelector('.cognito-body'); const inputEl=panel.querySelector('.cognito-input'); const sendBtn=panel.querySelector('.cognito-send');
function appendMessage(text, who){ const m=document.createElement('div'); m.className='cognito-msg '+(who==='user'?'user':'bot'); m.innerText=text; bodyEl.appendChild(m); bodyEl.scrollTop = bodyEl.scrollHeight; }
async function sendMessage(text){ appendMessage(text,'user'); inputEl.value=''; const payload={ messages: [{role:'user', content:text}], session_id: null }; const headers={'Content-Type':'application/json'}; if(CONFIG.apiKey) headers['x-api-key']=CONFIG.apiKey; try{ const res = await fetch(CONFIG.apiBase + '/chat',{ method:'POST', headers, body: JSON.stringify(payload) }); if(!res.ok){ appendMessage('Error: '+res.status+' '+await res.text(),'bot'); return; } const data = await res.json(); appendMessage(data.reply || JSON.stringify(data), 'bot'); } catch(err){ appendMessage('Network error: '+err.message,'bot'); } }
sendBtn.addEventListener('click', ()=>{ const v=inputEl.value.trim(); if(!v) return; sendMessage(v); });
inputEl.addEventListener('keydown', (e)=>{ if(e.key==='Enter'){ e.preventDefault(); sendBtn.click(); } });
window.CognitoWidget = { send: sendMessage, open: () => { panel.style.display='flex'; }, close: () => { panel.style.display='none'; } };
})();
WIDGET
# Provide default .env file
cat > .env <<'ENV'
# Cognito minimal .env - automatically used by setup script defaults if not provided
# Replace HF_API_TOKEN with a valid Hugging Face Inference API token to enable model responses.
HF_API_TOKEN=""
HF_MODEL_ID="gpt2"
COGNITO_API_KEY="cognito_default_key_please_change"
ALLOW_FETCH="false"
FETCH_BEARER_TOKEN=""
ENV
echo "==> Starting server (virtualenv active)."
echo "API key (for widget) default: cognito_default_key_please_change"
echo "If you have a Hugging Face token, export HF_API_TOKEN before running this script for real model responses."
echo "Example to run with token: HF_API_TOKEN='hf_xxx' ./setup_and_run.sh"
# Start uvicorn
# shellcheck disable=SC2086
uvicorn app:app --host 0.0.0.0 --port 8000 --reload