main.py 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457
  1. from contextlib import asynccontextmanager
  2. from bs4 import BeautifulSoup
  3. import json
  4. import markdown
  5. import time
  6. import os
  7. import sys
  8. import logging
  9. import aiohttp
  10. import requests
  11. from fastapi import FastAPI, Request, Depends, status
  12. from fastapi.staticfiles import StaticFiles
  13. from fastapi import HTTPException
  14. from fastapi.middleware.wsgi import WSGIMiddleware
  15. from fastapi.middleware.cors import CORSMiddleware
  16. from starlette.exceptions import HTTPException as StarletteHTTPException
  17. from starlette.middleware.base import BaseHTTPMiddleware
  18. from starlette.responses import StreamingResponse, Response
  19. from apps.ollama.main import app as ollama_app
  20. from apps.openai.main import app as openai_app
  21. from apps.litellm.main import (
  22. app as litellm_app,
  23. start_litellm_background,
  24. shutdown_litellm_background,
  25. )
  26. from apps.audio.main import app as audio_app
  27. from apps.images.main import app as images_app
  28. from apps.rag.main import app as rag_app
  29. from apps.web.main import app as webui_app
  30. import asyncio
  31. from pydantic import BaseModel
  32. from typing import List, Optional
  33. from apps.web.models.models import Models, ModelModel, ModelForm
  34. from utils.utils import get_admin_user
  35. from apps.rag.utils import rag_messages
  36. from config import (
  37. CONFIG_DATA,
  38. WEBUI_NAME,
  39. WEBUI_URL,
  40. WEBUI_AUTH,
  41. ENV,
  42. VERSION,
  43. CHANGELOG,
  44. FRONTEND_BUILD_DIR,
  45. CACHE_DIR,
  46. STATIC_DIR,
  47. ENABLE_LITELLM,
  48. ENABLE_MODEL_FILTER,
  49. MODEL_FILTER_LIST,
  50. GLOBAL_LOG_LEVEL,
  51. SRC_LOG_LEVELS,
  52. WEBHOOK_URL,
  53. ENABLE_ADMIN_EXPORT,
  54. AppConfig,
  55. )
  56. from constants import ERROR_MESSAGES
  57. logging.basicConfig(stream=sys.stdout, level=GLOBAL_LOG_LEVEL)
  58. log = logging.getLogger(__name__)
  59. log.setLevel(SRC_LOG_LEVELS["MAIN"])
  60. class SPAStaticFiles(StaticFiles):
  61. async def get_response(self, path: str, scope):
  62. try:
  63. return await super().get_response(path, scope)
  64. except (HTTPException, StarletteHTTPException) as ex:
  65. if ex.status_code == 404:
  66. return await super().get_response("index.html", scope)
  67. else:
  68. raise ex
  69. print(
  70. rf"""
  71. ___ __ __ _ _ _ ___
  72. / _ \ _ __ ___ _ __ \ \ / /__| |__ | | | |_ _|
  73. | | | | '_ \ / _ \ '_ \ \ \ /\ / / _ \ '_ \| | | || |
  74. | |_| | |_) | __/ | | | \ V V / __/ |_) | |_| || |
  75. \___/| .__/ \___|_| |_| \_/\_/ \___|_.__/ \___/|___|
  76. |_|
  77. v{VERSION} - building the best open-source AI user interface.
  78. https://github.com/open-webui/open-webui
  79. """
  80. )
  81. @asynccontextmanager
  82. async def lifespan(app: FastAPI):
  83. if ENABLE_LITELLM:
  84. asyncio.create_task(start_litellm_background())
  85. yield
  86. if ENABLE_LITELLM:
  87. await shutdown_litellm_background()
  88. app = FastAPI(
  89. docs_url="/docs" if ENV == "dev" else None, redoc_url=None, lifespan=lifespan
  90. )
  91. app.state.config = AppConfig()
  92. app.state.config.ENABLE_MODEL_FILTER = ENABLE_MODEL_FILTER
  93. app.state.config.MODEL_FILTER_LIST = MODEL_FILTER_LIST
  94. app.state.MODEL_CONFIG = [model.to_form() for model in Models.get_all_models()]
  95. app.state.config.WEBHOOK_URL = WEBHOOK_URL
  96. origins = ["*"]
  97. # Custom middleware to add security headers
  98. # class SecurityHeadersMiddleware(BaseHTTPMiddleware):
  99. # async def dispatch(self, request: Request, call_next):
  100. # response: Response = await call_next(request)
  101. # response.headers["Cross-Origin-Opener-Policy"] = "same-origin"
  102. # response.headers["Cross-Origin-Embedder-Policy"] = "require-corp"
  103. # return response
  104. # app.add_middleware(SecurityHeadersMiddleware)
  105. class RAGMiddleware(BaseHTTPMiddleware):
  106. async def dispatch(self, request: Request, call_next):
  107. return_citations = False
  108. if request.method == "POST" and (
  109. "/api/chat" in request.url.path or "/chat/completions" in request.url.path
  110. ):
  111. log.debug(f"request.url.path: {request.url.path}")
  112. # Read the original request body
  113. body = await request.body()
  114. # Decode body to string
  115. body_str = body.decode("utf-8")
  116. # Parse string to JSON
  117. data = json.loads(body_str) if body_str else {}
  118. return_citations = data.get("citations", False)
  119. if "citations" in data:
  120. del data["citations"]
  121. # Example: Add a new key-value pair or modify existing ones
  122. # data["modified"] = True # Example modification
  123. if "docs" in data:
  124. data = {**data}
  125. data["messages"], citations = rag_messages(
  126. docs=data["docs"],
  127. messages=data["messages"],
  128. template=rag_app.state.config.RAG_TEMPLATE,
  129. embedding_function=rag_app.state.EMBEDDING_FUNCTION,
  130. k=rag_app.state.config.TOP_K,
  131. reranking_function=rag_app.state.sentence_transformer_rf,
  132. r=rag_app.state.config.RELEVANCE_THRESHOLD,
  133. hybrid_search=rag_app.state.config.ENABLE_RAG_HYBRID_SEARCH,
  134. )
  135. del data["docs"]
  136. log.debug(
  137. f"data['messages']: {data['messages']}, citations: {citations}"
  138. )
  139. modified_body_bytes = json.dumps(data).encode("utf-8")
  140. # Replace the request body with the modified one
  141. request._body = modified_body_bytes
  142. # Set custom header to ensure content-length matches new body length
  143. request.headers.__dict__["_list"] = [
  144. (b"content-length", str(len(modified_body_bytes)).encode("utf-8")),
  145. *[
  146. (k, v)
  147. for k, v in request.headers.raw
  148. if k.lower() != b"content-length"
  149. ],
  150. ]
  151. response = await call_next(request)
  152. if return_citations:
  153. # Inject the citations into the response
  154. if isinstance(response, StreamingResponse):
  155. # If it's a streaming response, inject it as SSE event or NDJSON line
  156. content_type = response.headers.get("Content-Type")
  157. if "text/event-stream" in content_type:
  158. return StreamingResponse(
  159. self.openai_stream_wrapper(response.body_iterator, citations),
  160. )
  161. if "application/x-ndjson" in content_type:
  162. return StreamingResponse(
  163. self.ollama_stream_wrapper(response.body_iterator, citations),
  164. )
  165. return response
  166. async def _receive(self, body: bytes):
  167. return {"type": "http.request", "body": body, "more_body": False}
  168. async def openai_stream_wrapper(self, original_generator, citations):
  169. yield f"data: {json.dumps({'citations': citations})}\n\n"
  170. async for data in original_generator:
  171. yield data
  172. async def ollama_stream_wrapper(self, original_generator, citations):
  173. yield f"{json.dumps({'citations': citations})}\n"
  174. async for data in original_generator:
  175. yield data
  176. app.add_middleware(RAGMiddleware)
  177. app.add_middleware(
  178. CORSMiddleware,
  179. allow_origins=origins,
  180. allow_credentials=True,
  181. allow_methods=["*"],
  182. allow_headers=["*"],
  183. )
  184. @app.middleware("http")
  185. async def check_url(request: Request, call_next):
  186. start_time = int(time.time())
  187. response = await call_next(request)
  188. process_time = int(time.time()) - start_time
  189. response.headers["X-Process-Time"] = str(process_time)
  190. return response
  191. @app.middleware("http")
  192. async def update_embedding_function(request: Request, call_next):
  193. response = await call_next(request)
  194. if "/embedding/update" in request.url.path:
  195. webui_app.state.EMBEDDING_FUNCTION = rag_app.state.EMBEDDING_FUNCTION
  196. return response
  197. app.mount("/litellm/api", litellm_app)
  198. app.mount("/ollama", ollama_app)
  199. app.mount("/openai/api", openai_app)
  200. app.mount("/images/api/v1", images_app)
  201. app.mount("/audio/api/v1", audio_app)
  202. app.mount("/rag/api/v1", rag_app)
  203. app.mount("/api/v1", webui_app)
  204. webui_app.state.EMBEDDING_FUNCTION = rag_app.state.EMBEDDING_FUNCTION
  205. @app.get("/api/config")
  206. async def get_app_config():
  207. # Checking and Handling the Absence of 'ui' in CONFIG_DATA
  208. default_locale = "en-US"
  209. if "ui" in CONFIG_DATA:
  210. default_locale = CONFIG_DATA["ui"].get("default_locale", "en-US")
  211. # The Rest of the Function Now Uses the Variables Defined Above
  212. return {
  213. "status": True,
  214. "name": WEBUI_NAME,
  215. "version": VERSION,
  216. "auth": WEBUI_AUTH,
  217. "default_locale": default_locale,
  218. "images": images_app.state.config.ENABLED,
  219. "default_models": webui_app.state.config.DEFAULT_MODELS,
  220. "default_prompt_suggestions": webui_app.state.config.DEFAULT_PROMPT_SUGGESTIONS,
  221. "trusted_header_auth": bool(webui_app.state.AUTH_TRUSTED_EMAIL_HEADER),
  222. "admin_export_enabled": ENABLE_ADMIN_EXPORT,
  223. }
  224. @app.get("/api/config/model/filter")
  225. async def get_model_filter_config(user=Depends(get_admin_user)):
  226. return {
  227. "enabled": app.state.config.ENABLE_MODEL_FILTER,
  228. "models": app.state.config.MODEL_FILTER_LIST,
  229. }
  230. class ModelFilterConfigForm(BaseModel):
  231. enabled: bool
  232. models: List[str]
  233. @app.post("/api/config/model/filter")
  234. async def update_model_filter_config(
  235. form_data: ModelFilterConfigForm, user=Depends(get_admin_user)
  236. ):
  237. app.state.config.ENABLE_MODEL_FILTER = form_data.enabled
  238. app.state.config.MODEL_FILTER_LIST = form_data.models
  239. ollama_app.state.config.ENABLE_MODEL_FILTER = app.state.config.ENABLE_MODEL_FILTER
  240. ollama_app.state.config.MODEL_FILTER_LIST = app.state.config.MODEL_FILTER_LIST
  241. openai_app.state.config.ENABLE_MODEL_FILTER = app.state.config.ENABLE_MODEL_FILTER
  242. openai_app.state.config.MODEL_FILTER_LIST = app.state.config.MODEL_FILTER_LIST
  243. litellm_app.state.ENABLE_MODEL_FILTER = app.state.config.ENABLE_MODEL_FILTER
  244. litellm_app.state.MODEL_FILTER_LIST = app.state.config.MODEL_FILTER_LIST
  245. return {
  246. "enabled": app.state.config.ENABLE_MODEL_FILTER,
  247. "models": app.state.config.MODEL_FILTER_LIST,
  248. }
  249. class SetModelConfigForm(BaseModel):
  250. models: List[ModelForm]
  251. @app.post("/api/config/models")
  252. async def update_model_config(
  253. form_data: SetModelConfigForm, user=Depends(get_admin_user)
  254. ):
  255. if not Models.update_all_models(form_data.models):
  256. raise HTTPException(
  257. status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
  258. detail=ERROR_MESSAGES.DEFAULT("Failed to update model config"),
  259. )
  260. ollama_app.state.MODEL_CONFIG = [
  261. model for model in form_data.models if model.source == "ollama"
  262. ]
  263. openai_app.state.MODEL_CONFIG = [
  264. model for model in form_data.models if model.source == "openai"
  265. ]
  266. litellm_app.state.MODEL_CONFIG = [
  267. model for model in form_data.models if model.source == "litellm"
  268. ]
  269. app.state.MODEL_CONFIG = [model for model in form_data.models]
  270. return {"models": app.state.MODEL_CONFIG}
  271. @app.get("/api/config/models")
  272. async def get_model_config(user=Depends(get_admin_user)):
  273. return {"models": app.state.MODEL_CONFIG}
  274. @app.get("/api/webhook")
  275. async def get_webhook_url(user=Depends(get_admin_user)):
  276. return {
  277. "url": app.state.config.WEBHOOK_URL,
  278. }
  279. class UrlForm(BaseModel):
  280. url: str
  281. @app.post("/api/webhook")
  282. async def update_webhook_url(form_data: UrlForm, user=Depends(get_admin_user)):
  283. app.state.config.WEBHOOK_URL = form_data.url
  284. webui_app.state.WEBHOOK_URL = app.state.config.WEBHOOK_URL
  285. return {
  286. "url": app.state.config.WEBHOOK_URL,
  287. }
  288. @app.get("/api/version")
  289. async def get_app_config():
  290. return {
  291. "version": VERSION,
  292. }
  293. @app.get("/api/changelog")
  294. async def get_app_changelog():
  295. return {key: CHANGELOG[key] for idx, key in enumerate(CHANGELOG) if idx < 5}
  296. @app.get("/api/version/updates")
  297. async def get_app_latest_release_version():
  298. try:
  299. async with aiohttp.ClientSession() as session:
  300. async with session.get(
  301. "https://api.github.com/repos/open-webui/open-webui/releases/latest"
  302. ) as response:
  303. response.raise_for_status()
  304. data = await response.json()
  305. latest_version = data["tag_name"]
  306. return {"current": VERSION, "latest": latest_version[1:]}
  307. except aiohttp.ClientError as e:
  308. raise HTTPException(
  309. status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
  310. detail=ERROR_MESSAGES.RATE_LIMIT_EXCEEDED,
  311. )
  312. @app.get("/manifest.json")
  313. async def get_manifest_json():
  314. return {
  315. "name": WEBUI_NAME,
  316. "short_name": WEBUI_NAME,
  317. "start_url": "/",
  318. "display": "standalone",
  319. "background_color": "#343541",
  320. "theme_color": "#343541",
  321. "orientation": "portrait-primary",
  322. "icons": [{"src": "/static/logo.png", "type": "image/png", "sizes": "500x500"}],
  323. }
  324. @app.get("/opensearch.xml")
  325. async def get_opensearch_xml():
  326. xml_content = rf"""
  327. <OpenSearchDescription xmlns="http://a9.com/-/spec/opensearch/1.1/" xmlns:moz="http://www.mozilla.org/2006/browser/search/">
  328. <ShortName>{WEBUI_NAME}</ShortName>
  329. <Description>Search {WEBUI_NAME}</Description>
  330. <InputEncoding>UTF-8</InputEncoding>
  331. <Image width="16" height="16" type="image/x-icon">{WEBUI_URL}/favicon.png</Image>
  332. <Url type="text/html" method="get" template="{WEBUI_URL}/?q={"{searchTerms}"}"/>
  333. <moz:SearchForm>{WEBUI_URL}</moz:SearchForm>
  334. </OpenSearchDescription>
  335. """
  336. return Response(content=xml_content, media_type="application/xml")
  337. @app.get("/health")
  338. async def healthcheck():
  339. return {"status": True}
  340. app.mount("/static", StaticFiles(directory=STATIC_DIR), name="static")
  341. app.mount("/cache", StaticFiles(directory=CACHE_DIR), name="cache")
  342. if os.path.exists(FRONTEND_BUILD_DIR):
  343. app.mount(
  344. "/",
  345. SPAStaticFiles(directory=FRONTEND_BUILD_DIR, html=True),
  346. name="spa-static-files",
  347. )
  348. else:
  349. log.warning(
  350. f"Frontend build directory not found at '{FRONTEND_BUILD_DIR}'. Serving API only."
  351. )