| import os
|
| import time
|
| from datetime import datetime, timezone
|
| from html import escape
|
|
|
| from fastapi import Depends, FastAPI, HTTPException
|
| from fastapi.middleware.cors import CORSMiddleware
|
| from fastapi.responses import HTMLResponse
|
|
|
| from .adapter import build_visualization_from_intent
|
| from .config import settings
|
| from .db import DatabaseHandler
|
| from .knowledge import KB_CONFIG, KnowledgeManager
|
| from .models import AiQueryRequest, AiQueryResponse
|
| from .security import require_api_key
|
| from .ai_engine import AIEngine
|
|
|
|
|
| app = FastAPI(title="Pravah AI Backend", version="1.0.0")
|
|
|
| origins = [o.strip() for o in settings.allowed_origins.split(",") if o.strip()]
|
| if origins == ["*"]:
|
| allow_origins = ["*"]
|
| allow_credentials = False
|
| else:
|
| allow_origins = origins
|
| allow_credentials = True
|
|
|
| app.add_middleware(
|
| CORSMiddleware,
|
| allow_origins=allow_origins,
|
| allow_credentials=allow_credentials,
|
| allow_methods=["GET", "POST", "OPTIONS"],
|
| allow_headers=["*"]
|
| if not settings.backend_api_key
|
| else ["Content-Type", "Authorization", "X-API-Key"],
|
| )
|
|
|
|
|
| db = DatabaseHandler()
|
| kb = KnowledgeManager(KB_CONFIG)
|
| ai = AIEngine(db, kb)
|
|
|
| _LOGS: list[dict] = []
|
| _MAX_LOGS = 500
|
|
|
|
|
| def _append_log(entry: dict) -> None:
|
| _LOGS.append(entry)
|
| if len(_LOGS) > _MAX_LOGS:
|
| del _LOGS[0 : max(0, len(_LOGS) - _MAX_LOGS)]
|
|
|
|
|
| @app.get("/health")
|
| def health():
|
| return {
|
| "ok": True,
|
| "build": os.environ.get("SPACE_REVISION")
|
| or os.environ.get("HF_SPACE_REVISION")
|
| or os.environ.get("GIT_SHA")
|
| or "local",
|
| "features": {
|
| "autoInsights": True,
|
| "sqlIdentifierNormalization": True,
|
| "rag": True,
|
| },
|
| "dbConfigured": db.is_configured(),
|
| "db": db.healthcheck(),
|
| "groqConfigured": ai.is_configured(),
|
| }
|
|
|
|
|
| @app.get("/admin/logs", response_class=HTMLResponse, dependencies=[Depends(require_api_key)])
|
| def admin_logs():
|
| rows = list(reversed(_LOGS))
|
|
|
| def td(v) -> str:
|
| if v is None:
|
| return "<td class='muted'>—</td>"
|
| return f"<td>{escape(str(v))}</td>"
|
|
|
| html = """
|
| <html>
|
| <head>
|
| <meta charset='utf-8' />
|
| <meta name='viewport' content='width=device-width, initial-scale=1' />
|
| <title>Pravah Backend Logs</title>
|
| <style>
|
| body{font-family:ui-sans-serif,system-ui,-apple-system,Segoe UI,Roboto,Arial; background:#0b1220; color:#e5e7eb; padding:16px;}
|
| .card{background:#0f172a; border:1px solid #1f2937; border-radius:12px; padding:14px;}
|
| h1{font-size:18px; margin:0 0 10px 0;}
|
| .sub{color:#9ca3af; font-size:12px; margin-bottom:12px;}
|
| table{width:100%; border-collapse:collapse; font-size:12px;}
|
| th,td{border-bottom:1px solid #1f2937; padding:8px; vertical-align:top;}
|
| th{color:#cbd5e1; text-align:left; font-weight:700; position:sticky; top:0; background:#0f172a;}
|
| .ok{color:#34d399; font-weight:700;}
|
| .err{color:#fb7185; font-weight:700;}
|
| .muted{color:#94a3b8;}
|
| code{color:#a5b4fc;}
|
| .pill{display:inline-block; padding:2px 8px; border-radius:999px; background:#111827; border:1px solid #1f2937; font-size:11px; color:#cbd5e1;}
|
| </style>
|
| </head>
|
| <body>
|
| <div class='card'>
|
| <h1>Pravah Backend Logs</h1>
|
| <div class='sub'>Showing last <span class='pill'>""" + str(len(rows)) + """</span> requests (max """ + str(_MAX_LOGS) + """ in memory). Use <code>/admin/logs.json</code> for raw data.</div>
|
| <table>
|
| <thead>
|
| <tr>
|
| <th>Time (UTC)</th>
|
| <th>Status</th>
|
| <th>Intent</th>
|
| <th>Model</th>
|
| <th>Latency (s)</th>
|
| <th>Tokens</th>
|
| <th>Prompt</th>
|
| <th>SQL</th>
|
| <th>Error</th>
|
| </tr>
|
| </thead>
|
| <tbody>
|
| """
|
|
|
| for r in rows:
|
| status = r.get("status")
|
| status_html = "<span class='ok'>200</span>" if status == 200 else f"<span class='err'>{escape(str(status))}</span>"
|
| html += "<tr>"
|
| html += td(r.get("ts"))
|
| html += f"<td>{status_html}</td>"
|
| html += td(r.get("intent"))
|
| html += td(r.get("model"))
|
| html += td(r.get("latency"))
|
| tok = r.get("tokens")
|
| if tok is None:
|
| html += "<td class='muted'>—</td>"
|
| else:
|
| html += f"<td><span class='pill'>{escape(str(tok))}</span></td>"
|
| html += td(r.get("prompt"))
|
| sql = r.get("sql")
|
| if sql:
|
| html += f"<td><code>{escape(str(sql))}</code></td>"
|
| else:
|
| html += "<td class='muted'>—</td>"
|
| err = r.get("error")
|
| if err:
|
| html += f"<td class='err'>{escape(str(err))}</td>"
|
| else:
|
| html += "<td class='muted'>—</td>"
|
| html += "</tr>"
|
|
|
| html += """
|
| </tbody>
|
| </table>
|
| </div>
|
| </body>
|
| </html>
|
| """
|
| return HTMLResponse(content=html)
|
|
|
|
|
| @app.get("/admin/logs.json", dependencies=[Depends(require_api_key)])
|
| def admin_logs_json():
|
| return {"items": _LOGS[-_MAX_LOGS:], "count": len(_LOGS)}
|
|
|
|
|
| @app.post("/api/ai/query", response_model=AiQueryResponse, dependencies=[Depends(require_api_key)])
|
| def ai_query(payload: AiQueryRequest):
|
| if not ai.is_configured():
|
| raise HTTPException(status_code=503, detail="Groq is not configured. Set GROQ_API_KEY.")
|
|
|
| started = time.time()
|
| sql = None
|
| intent = None
|
| model = None
|
| tokens = None
|
| error = None
|
| status = 200
|
|
|
| try:
|
| plan, metrics = ai.process_query(payload.prompt, history=[])
|
| model = metrics.get("model")
|
| tokens = metrics.get("tokens")
|
|
|
| intent = plan.get("intent", "chat")
|
| reply = plan.get("message", "")
|
| sql = plan.get("sql")
|
| chart_config = plan.get("chart_config")
|
|
|
| df = None
|
| if intent in ["sql", "analytics"] and isinstance(sql, str) and sql.strip():
|
| if not db.is_configured():
|
| return AiQueryResponse(
|
| answer=f"{reply}\n\nDatabase not configured. Set POSTGRES_* settings.",
|
| visualization=None,
|
| )
|
|
|
| df, err = db.execute_query(sql)
|
| if err:
|
| fixed_sql = ai.repair_sql(payload.prompt, sql, err)
|
| if fixed_sql and fixed_sql.strip() and fixed_sql.strip() != sql.strip():
|
| sql = fixed_sql
|
| df, err2 = db.execute_query(sql)
|
| if err2:
|
| return AiQueryResponse(
|
| answer=f"{reply}\n\nSQL Error: {err2}",
|
| visualization=None,
|
| )
|
| else:
|
| return AiQueryResponse(answer=f"{reply}\n\nSQL Error: {err}", visualization=None)
|
|
|
| answer, visualization = build_visualization_from_intent(intent, reply, df, sql, chart_config)
|
| return AiQueryResponse(answer=answer, visualization=visualization)
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| status = 500
|
| error = str(e)
|
| raise
|
| finally:
|
| latency = round(time.time() - started, 3)
|
| _append_log(
|
| {
|
| "ts": datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S"),
|
| "status": status,
|
| "intent": intent,
|
| "model": model,
|
| "latency": latency,
|
| "tokens": tokens,
|
| "prompt": (payload.prompt or "")[:2000],
|
| "sql": (sql or "")[:4000] if isinstance(sql, str) else None,
|
| "error": error,
|
| }
|
| )
|
|
|
