553 lines
19 KiB
Python
553 lines
19 KiB
Python
import asyncio
|
|
from fastapi import FastAPI, Depends, HTTPException
|
|
from fastapi.middleware.cors import CORSMiddleware
|
|
from fastapi.responses import StreamingResponse
|
|
from sqlalchemy.orm import Session
|
|
from typing import Any, List
|
|
import re
|
|
import html
|
|
import json
|
|
import base64
|
|
from . import models, schemas
|
|
from .database import Base, engine, SessionLocal, ensure_sources_column
|
|
from .local_rag import router as local_rag_router
|
|
from .ollama_admin import inspect_ollama_startup, prepare_startup_models, pull_local_model, start_local_ollama
|
|
from .ollama_client import (
|
|
list_model_catalog as ollama_list_model_catalog,
|
|
chat as ollama_chat,
|
|
chat_stream as ollama_chat_stream,
|
|
show_model as ollama_show_model,
|
|
supports_vision as ollama_supports_vision,
|
|
)
|
|
from .whisper_admin import DEFAULT_WHISPER_MODEL, list_whisper_models, transcribe_audio_bytes
|
|
from .websearch import enrich_prompt
|
|
|
|
# Create tables + ensure migration
|
|
Base.metadata.create_all(bind=engine)
|
|
ensure_sources_column(engine)
|
|
|
|
app = FastAPI(title="LLM Desktop Backend", version="0.1.0" )
|
|
|
|
|
|
def sanitize_chat_title(title: str) -> str:
|
|
cleaned_title = html.unescape(title or "")
|
|
cleaned_title = re.sub(r'<think(?:ing)?>.*?</think(?:ing)?>', '', cleaned_title, flags=re.DOTALL | re.IGNORECASE)
|
|
cleaned_title = cleaned_title.strip()
|
|
|
|
previous_title = None
|
|
while cleaned_title and cleaned_title != previous_title:
|
|
previous_title = cleaned_title
|
|
cleaned_title = re.sub(r'^\s*#+\s*', '', cleaned_title)
|
|
cleaned_title = re.sub(r'^\s*\*{1,2}\s*', '', cleaned_title)
|
|
cleaned_title = re.sub(r'\s*\*{1,2}\s*$', '', cleaned_title)
|
|
cleaned_title = cleaned_title.strip()
|
|
|
|
cleaned_title = re.sub(r'\s+', ' ', cleaned_title)
|
|
return cleaned_title.strip()
|
|
|
|
# CORS (dev-friendly; tighten later)
|
|
app.add_middleware(
|
|
CORSMiddleware,
|
|
allow_origins=["*"],
|
|
allow_credentials=True,
|
|
allow_methods=["*"],
|
|
allow_headers=["*"],
|
|
)
|
|
app.include_router(local_rag_router)
|
|
|
|
_IMAGE_DATA_URL_RE = re.compile(r"^data:(image\/[a-z0-9.+-]+);base64,([a-z0-9+/=\s]+)$", re.IGNORECASE)
|
|
|
|
|
|
def _attachment_field(item: Any, field: str) -> Any:
|
|
if isinstance(item, dict):
|
|
return item.get(field)
|
|
return getattr(item, field, None)
|
|
|
|
|
|
def _normalize_image_attachments(items: Any) -> List[dict]:
|
|
cleaned: List[dict] = []
|
|
for item in items or []:
|
|
data_url = str(_attachment_field(item, "data_url") or "").strip()
|
|
name = str(_attachment_field(item, "name") or "image").strip() or "image"
|
|
mime_type = str(_attachment_field(item, "mime_type") or "").strip().lower()
|
|
match = _IMAGE_DATA_URL_RE.match(data_url)
|
|
if not match:
|
|
continue
|
|
|
|
detected_mime = match.group(1).lower()
|
|
payload = re.sub(r"\s+", "", match.group(2))
|
|
if mime_type and not mime_type.startswith("image/"):
|
|
continue
|
|
|
|
try:
|
|
base64.b64decode(payload, validate=True)
|
|
except Exception:
|
|
continue
|
|
|
|
cleaned.append({
|
|
"name": name[:255],
|
|
"mime_type": mime_type or detected_mime,
|
|
"data_url": f"data:{detected_mime};base64,{payload}",
|
|
})
|
|
return cleaned
|
|
|
|
|
|
def _load_message_attachments(raw_value: Any) -> List[dict]:
|
|
if isinstance(raw_value, str):
|
|
try:
|
|
parsed = json.loads(raw_value or "[]")
|
|
except Exception:
|
|
parsed = []
|
|
else:
|
|
parsed = raw_value
|
|
return _normalize_image_attachments(parsed)
|
|
|
|
|
|
def _attachments_to_ollama_images(attachments: List[dict]) -> List[str]:
|
|
images: List[str] = []
|
|
for attachment in attachments:
|
|
match = _IMAGE_DATA_URL_RE.match(str(attachment.get("data_url") or "").strip())
|
|
if not match:
|
|
continue
|
|
images.append(re.sub(r"\s+", "", match.group(2)))
|
|
return images
|
|
|
|
|
|
def _row_to_history_message(row: models.ChatMessage) -> dict:
|
|
sources = []
|
|
try:
|
|
if getattr(row, "sources_json", None):
|
|
sources = json.loads(row.sources_json or "[]")
|
|
except Exception:
|
|
sources = []
|
|
|
|
attachments = _load_message_attachments(getattr(row, "attachments_json", None))
|
|
payload = {"role": row.role, "content": row.content, "sources": sources}
|
|
if attachments:
|
|
payload["attachments"] = attachments
|
|
return payload
|
|
|
|
|
|
def _row_to_ollama_message(row: models.ChatMessage) -> dict:
|
|
message = {"role": row.role, "content": row.content}
|
|
attachments = _load_message_attachments(getattr(row, "attachments_json", None))
|
|
if attachments:
|
|
message["images"] = _attachments_to_ollama_images(attachments)
|
|
return message
|
|
|
|
def get_db():
|
|
db = SessionLocal()
|
|
try:
|
|
yield db
|
|
finally:
|
|
db.close()
|
|
|
|
@app.get("/health")
|
|
def health():
|
|
return {"ok": True}
|
|
|
|
|
|
@app.post("/audio/transcribe", response_model=schemas.AudioTranscriptionResponse)
|
|
async def transcribe_audio_route(req: schemas.AudioTranscriptionRequest):
|
|
mime_type = str(req.mime_type or "").split(";", 1)[0].strip().lower()
|
|
if not mime_type.startswith("audio/"):
|
|
raise HTTPException(status_code=400, detail="An audio mime type is required.")
|
|
|
|
payload = re.sub(r"\s+", "", str(req.audio_base64 or ""))
|
|
if not payload:
|
|
raise HTTPException(status_code=400, detail="Audio payload is required.")
|
|
|
|
try:
|
|
audio_bytes = base64.b64decode(payload, validate=True)
|
|
except Exception as exc:
|
|
raise HTTPException(status_code=400, detail="Invalid base64 audio payload.") from exc
|
|
|
|
try:
|
|
result = await asyncio.to_thread(
|
|
transcribe_audio_bytes,
|
|
audio_bytes,
|
|
mime_type,
|
|
req.model or DEFAULT_WHISPER_MODEL,
|
|
req.language,
|
|
)
|
|
except RuntimeError as exc:
|
|
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
|
except Exception as exc:
|
|
raise HTTPException(status_code=500, detail=f"Audio transcription failed: {exc}") from exc
|
|
|
|
return {
|
|
"text": str(result.get("text") or "").strip(),
|
|
"language": str(result.get("language") or "").strip() or None,
|
|
"model": str(result.get("model") or req.model or DEFAULT_WHISPER_MODEL),
|
|
}
|
|
|
|
@app.get("/models")
|
|
async def get_models():
|
|
try:
|
|
ollama_data, whisper_data = await asyncio.gather(
|
|
ollama_list_model_catalog(),
|
|
asyncio.to_thread(list_whisper_models),
|
|
)
|
|
return {
|
|
**ollama_data,
|
|
"whisper_models": whisper_data.get("models", []),
|
|
"whisper_error": whisper_data.get("error", ""),
|
|
}
|
|
except Exception as e:
|
|
raise HTTPException(status_code=502, detail=f"Ollama not available: {e}")
|
|
|
|
|
|
@app.get("/models/capabilities")
|
|
async def get_model_capabilities(name: str):
|
|
model_name = str(name or "").strip()
|
|
if not model_name:
|
|
raise HTTPException(status_code=400, detail="Model name is required.")
|
|
|
|
try:
|
|
model_data = await ollama_show_model(model_name)
|
|
except Exception as e:
|
|
raise HTTPException(status_code=502, detail=f"Ollama not available: {e}")
|
|
|
|
capabilities = [
|
|
str(item).strip()
|
|
for item in (model_data.get("capabilities") or [])
|
|
if str(item).strip()
|
|
]
|
|
return {
|
|
"name": model_name,
|
|
"capabilities": capabilities,
|
|
"supports_vision": ollama_supports_vision(model_data),
|
|
}
|
|
|
|
|
|
@app.get("/ollama/startup-status")
|
|
async def ollama_startup_status():
|
|
return await inspect_ollama_startup()
|
|
|
|
|
|
@app.post("/ollama/start")
|
|
async def ollama_start_route():
|
|
try:
|
|
return await start_local_ollama()
|
|
except FileNotFoundError as exc:
|
|
raise HTTPException(status_code=404, detail=str(exc)) from exc
|
|
except RuntimeError as exc:
|
|
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
|
|
|
|
|
@app.post("/ollama/pull")
|
|
async def ollama_pull_route(req: schemas.OllamaPullRequest):
|
|
try:
|
|
return await pull_local_model(req.model)
|
|
except FileNotFoundError as exc:
|
|
raise HTTPException(status_code=404, detail=str(exc)) from exc
|
|
except RuntimeError as exc:
|
|
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
|
|
|
|
|
@app.post("/startup/prepare-models")
|
|
async def startup_prepare_models_route():
|
|
try:
|
|
return await prepare_startup_models()
|
|
except FileNotFoundError as exc:
|
|
raise HTTPException(status_code=404, detail=str(exc)) from exc
|
|
except RuntimeError as exc:
|
|
raise HTTPException(status_code=400, detail=str(exc)) from exc
|
|
|
|
@app.get("/sessions", response_model=schemas.SessionsResponse)
|
|
def get_sessions(db: Session = Depends(get_db)):
|
|
sessions = db.query(models.ChatSession).order_by(models.ChatSession.created_at.desc()).all()
|
|
return {
|
|
"sessions": [
|
|
{
|
|
"id": session.id,
|
|
"session_id": session.session_id,
|
|
"name": sanitize_chat_title(session.name),
|
|
"created_at": session.created_at,
|
|
}
|
|
for session in sessions
|
|
]
|
|
}
|
|
|
|
@app.post("/sessions", response_model=schemas.ChatSession)
|
|
def create_session(req: schemas.CreateSessionRequest, db: Session = Depends(get_db)):
|
|
new_session = models.ChatSession(session_id=req.session_id)
|
|
db.add(new_session)
|
|
db.commit()
|
|
db.refresh(new_session)
|
|
return new_session
|
|
|
|
@app.get("/history", response_model=schemas.HistoryResponse)
|
|
def history(session_id: str, db: Session = Depends(get_db)):
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == session_id).first()
|
|
if not session:
|
|
return {"messages": []}
|
|
rows = (
|
|
db.query(models.ChatMessage)
|
|
.filter(models.ChatMessage.session_pk == session.id)
|
|
.order_by(models.ChatMessage.created_at.asc())
|
|
.all()
|
|
)
|
|
msgs = [_row_to_history_message(r) for r in rows]
|
|
return {"messages": msgs}
|
|
|
|
@app.post("/chat")
|
|
async def chat(req: schemas.ChatRequest, db: Session = Depends(get_db)):
|
|
# Find or create session
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == req.session_id).first()
|
|
if not session:
|
|
session = models.ChatSession(session_id=req.session_id)
|
|
db.add(session)
|
|
db.commit()
|
|
db.refresh(session)
|
|
|
|
# Store the BASE user prompt
|
|
user_attachments = _normalize_image_attachments(req.attachments)
|
|
user_row = models.ChatMessage(
|
|
session_pk=session.id,
|
|
role='user',
|
|
content=req.message,
|
|
attachments_json=json.dumps(user_attachments or []),
|
|
)
|
|
db.add(user_row)
|
|
db.commit()
|
|
|
|
# Build minimal context (last 20)
|
|
last_msgs = (
|
|
db.query(models.ChatMessage)
|
|
.filter(models.ChatMessage.session_pk == session.id)
|
|
.order_by(models.ChatMessage.created_at.asc())
|
|
.all()[-20:]
|
|
)
|
|
messages = [_row_to_ollama_message(m) for m in last_msgs]
|
|
|
|
# Patch last user with enriched_message only for LLM call
|
|
if req.enriched_message:
|
|
for i in range(len(messages) - 1, -1, -1):
|
|
if messages[i].get("role") == "user":
|
|
messages = messages.copy()
|
|
messages[i] = {**messages[i], "content": req.enriched_message}
|
|
break
|
|
|
|
# Sources to persist with the assistant reply
|
|
sources = req.sources or []
|
|
|
|
if req.stream:
|
|
async def stream_generator():
|
|
full_reply = ""
|
|
try:
|
|
async for chunk in ollama_chat_stream(req.model, messages):
|
|
full_reply += chunk
|
|
yield chunk
|
|
except Exception as e:
|
|
yield f"Ollama error: {e}"
|
|
|
|
# Persist assistant reply (include sources_json)
|
|
as_row = models.ChatMessage(
|
|
session_pk=session.id, role='assistant', content=full_reply,
|
|
sources_json=json.dumps(sources or [])
|
|
)
|
|
db.add(as_row)
|
|
db.commit()
|
|
|
|
return StreamingResponse(stream_generator(), media_type="text/plain")
|
|
else:
|
|
try:
|
|
reply = await ollama_chat(req.model, messages)
|
|
except Exception as e:
|
|
raise HTTPException(status_code=502, detail=f"Ollama error: {e}")
|
|
|
|
as_row = models.ChatMessage(
|
|
session_pk=session.id, role='assistant', content=reply,
|
|
sources_json=json.dumps(sources or [])
|
|
)
|
|
db.add(as_row)
|
|
db.commit()
|
|
return {"reply": reply}
|
|
|
|
@app.post("/generate-title", response_model=schemas.GenerateTitleResponse)
|
|
async def generate_title(req: schemas.GenerateTitleRequest, db: Session = Depends(get_db)):
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == req.session_id).first()
|
|
if not session:
|
|
raise HTTPException(status_code=404, detail="Session not found")
|
|
|
|
prompt = f"Generate a very short, concise title (5 words or less) for a chat conversation that begins with this user message: \"{req.message}\". Do not use quotation marks in the title."
|
|
|
|
try:
|
|
title = await ollama_chat(req.model, [{"role": "user", "content": prompt}])
|
|
except Exception as e:
|
|
raise HTTPException(status_code=502, detail=f"Ollama error: {e}")
|
|
|
|
print(f"Original title from LLM: {title}") # Debugging line to see the raw title
|
|
|
|
cleaned_title = sanitize_chat_title(title)
|
|
|
|
print(f"Cleaned title before saving: {cleaned_title}") # Debugging line to see the cleaned title
|
|
|
|
session.name = cleaned_title
|
|
db.commit()
|
|
|
|
return {"title": cleaned_title}
|
|
|
|
@app.delete("/sessions/{session_id}")
|
|
def delete_session(session_id: str, db: Session = Depends(get_db)):
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == session_id).first()
|
|
if not session:
|
|
raise HTTPException(status_code=404, detail="Session not found")
|
|
|
|
# Delete associated messages
|
|
db.query(models.ChatMessage).filter(models.ChatMessage.session_pk == session.id).delete()
|
|
|
|
db.delete(session)
|
|
db.commit()
|
|
return {"ok": True}
|
|
|
|
@app.put("/sessions/{session_id}/rename")
|
|
def rename_session(session_id: str, req: schemas.GenerateTitleResponse, db: Session = Depends(get_db)):
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == session_id).first()
|
|
if not session:
|
|
raise HTTPException(status_code=404, detail="Session not found")
|
|
|
|
session.name = sanitize_chat_title(req.title)
|
|
db.commit()
|
|
return {"ok": True}
|
|
|
|
@app.put("/sessions/{session_id}/messages/{index}")
|
|
def update_user_message(session_id: str, index: int, req: schemas.EditMessageRequest, db: Session = Depends(get_db)):
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == session_id).first()
|
|
if not session:
|
|
raise HTTPException(status_code=404, detail="Session not found")
|
|
|
|
msgs = (
|
|
db.query(models.ChatMessage)
|
|
.filter(models.ChatMessage.session_pk == session.id)
|
|
.order_by(models.ChatMessage.created_at.asc())
|
|
.all()
|
|
)
|
|
|
|
if index < 0 or index >= len(msgs):
|
|
raise HTTPException(status_code=404, detail="Message index out of range")
|
|
|
|
# Only user messages can be edited per spec
|
|
if msgs[index].role != "user":
|
|
raise HTTPException(status_code=400, detail="Only user messages can be edited")
|
|
|
|
# Update the content
|
|
msgs[index].content = req.message
|
|
|
|
# Drop everything after the edited message
|
|
for m in msgs[index + 1:]:
|
|
db.delete(m)
|
|
|
|
db.commit()
|
|
return {"ok": True}
|
|
|
|
# ADD or REPLACE this whole function
|
|
@app.post("/sessions/{session_id}/regenerate")
|
|
async def regenerate(session_id: str, req: schemas.RegenerateRequest, db: Session = Depends(get_db)):
|
|
idx = req.index
|
|
model = req.model
|
|
stream = bool(req.stream)
|
|
sources = req.sources or []
|
|
|
|
session = db.query(models.ChatSession).filter(models.ChatSession.session_id == session_id).first()
|
|
if not session:
|
|
raise HTTPException(status_code=404, detail="Session not found")
|
|
|
|
msgs = (
|
|
db.query(models.ChatMessage)
|
|
.filter(models.ChatMessage.session_pk == session.id)
|
|
.order_by(models.ChatMessage.created_at.asc())
|
|
.all()
|
|
)
|
|
if idx < 0 or idx >= len(msgs):
|
|
raise HTTPException(status_code=400, detail="Invalid message index")
|
|
|
|
# last user idx at/before idx
|
|
last_user_idx = idx
|
|
for i in range(idx, -1, -1):
|
|
if msgs[i].role == "user":
|
|
last_user_idx = i
|
|
break
|
|
|
|
# prune after that user
|
|
if last_user_idx < len(msgs) - 1:
|
|
for m in msgs[last_user_idx + 1:]:
|
|
db.delete(m)
|
|
db.commit()
|
|
|
|
conversation = [_row_to_ollama_message(m) for m in msgs[: last_user_idx + 1]]
|
|
|
|
if req.enriched_message:
|
|
for j in range(len(conversation) - 1, -1, -1):
|
|
if conversation[j].get("role") == "user":
|
|
conversation = conversation.copy()
|
|
conversation[j] = {**conversation[j], "content": req.enriched_message}
|
|
break
|
|
|
|
session_pk = session.id
|
|
|
|
if stream:
|
|
async def stream_generator():
|
|
full_reply = ""
|
|
try:
|
|
async for chunk in ollama_chat_stream(model, conversation):
|
|
full_reply += chunk
|
|
yield chunk
|
|
except Exception as e:
|
|
yield f"Ollama error: {e}"
|
|
# persist (with sources)
|
|
try:
|
|
db_sess = SessionLocal()
|
|
db_sess.add(models.ChatMessage(
|
|
session_pk=session_pk, role="assistant", content=full_reply,
|
|
sources_json=json.dumps(sources or [])
|
|
))
|
|
db_sess.commit()
|
|
finally:
|
|
try:
|
|
db_sess.close()
|
|
except Exception:
|
|
pass
|
|
|
|
return StreamingResponse(stream_generator(), media_type="text/plain")
|
|
|
|
try:
|
|
reply = await ollama_chat(model, conversation)
|
|
except Exception as e:
|
|
raise HTTPException(status_code=502, detail=f"Ollama error: {e}")
|
|
|
|
db.add(models.ChatMessage(
|
|
session_pk=session_pk, role="assistant", content=reply,
|
|
sources_json=json.dumps(sources or [])
|
|
))
|
|
db.commit()
|
|
return {"reply": reply}
|
|
|
|
|
|
# -----------------------------------------------------------------------------
|
|
# Web search enrichment endpoint
|
|
@app.post("/websearch", response_model=schemas.WebSearchResponse)
|
|
async def websearch_route(req: schemas.WebSearchRequest):
|
|
"""
|
|
Return an enriched prompt (with citations) for a given user prompt.
|
|
Optionally uses the last `history_limit` turns from `req.messages`.
|
|
"""
|
|
try:
|
|
messages = (req.messages or [])[-int(req.history_limit or 8):]
|
|
enriched, sources = await enrich_prompt(
|
|
user_prompt=req.prompt,
|
|
model=req.model,
|
|
messages=[{"role": m.role, "content": m.content} for m in messages],
|
|
searx_url=req.searx_url,
|
|
engines=req.engines,
|
|
)
|
|
context_block = ""
|
|
if "<websearch_context>" in enriched:
|
|
context_block = enriched[enriched.index("<websearch_context>"):].strip()
|
|
return {"enriched_prompt": enriched, "sources": sources, "context_block": context_block}
|
|
except Exception:
|
|
return {"enriched_prompt": req.prompt, "sources": [], "context_block": ""}
|
|
|
|
# To run standalone: python -m uvicorn backend.main:app --host 127.0.0.1 --port 8000
|