feat(agent): Gradio-powered LangGraph agent chat with streaming, tool calls, file upload, conversation persistence

- Gradio 5.50.0 ChatInterface with type='messages' streaming
- LangGraph create_react_agent with InMemorySaver checkpointer
- 4 @tool functions: search_dashboards, get_health_summary, list_environments, get_task_status
- Structured ChatMessage metadata (7 discriminator types: stream_token, tool_start/end/error, confirm_required, confirm_resolved, error)
- HITL resume via second submit() with interrupt_before/Command
- Dual-identity RBAC: service JWT + user JWT for tool calls
- File upload (10 MB limit, pdfplumber/xlsx/JSON parser)
- Conversation persistence via POST /api/agent/conversations/save
- REST API: list, history, archive conversations; multi-tab gate; LLM config
- LLM provider selection via Admin -> LLM Settings (assistant_planner_provider)
- Svelte 5 AgentChatModel with stream event queue, dedup, stream_status watcher
- MarkdownRenderer using svelte-markdown with semantic Tailwind tokens
- ToolCallCard (3 states: executing/completed/failed)
- ConversationList with search, date grouping, infinite scroll
- ConnectionIndicator with Gradio health status
- /agent route with two-column layout
- Vite proxy /api/agent/gradio -> Gradio SSE
- Fixed: not_() SQLAlchemy operator, route collision with _admin_routes
- Fixed: conversation_id -> id normalization, .pyc cache staleness
- Fixed: event.data array parsing (Gradio returns [jsonStr, null])
- Requirements pinned: gradio==5.50.0, pydantic>=2.7,<=2.12.3
This commit is contained in:
2026-06-10 10:27:19 +03:00
parent 2222261157
commit f87ebf5d4b
28 changed files with 2863 additions and 140 deletions

291
backend/src/agent/app.py Normal file
View File

@@ -0,0 +1,291 @@
# backend/src/agent/app.py
# #region AgentChat.GradioApp [C:4] [TYPE Module] [SEMANTICS agent-chat,gradio,app]
# @defgroup AgentChat Gradio ChatInterface wrapping LangGraph agent. Streaming via submit(), HITL via interrupt().
# @PRE JWT_SECRET env var set. Shared with FastAPI for stateless validation.
# @POST Agent streams tokens via Gradio yield; audit logged via LoggingMiddleware.
# @SIDE_EFFECT Calls LLM, invokes tools via FastAPI REST, writes checkpoints to PostgreSQL.
# @RELATION DEPENDS_ON -> [AgentChat.LangGraph.Setup]
# @RELATION DEPENDS_ON -> [AgentChat.Context]
# @RELATION DEPENDS_ON -> [AgentChat.Tools]
# @RELATION DEPENDS_ON -> [AgentChat.Document.Parser]
from collections.abc import AsyncGenerator
import json
import os
import uuid
import gradio as gr
import httpx
import jwt
from langchain_core.exceptions import OutputParserException
from langchain_core.messages import HumanMessage
from langgraph.types import Command
from src.agent.context import set_user_jwt
from src.agent.document_parser import parse_upload
from src.agent.langgraph_setup import create_agent
from src.agent.middleware import log_tool_event
from src.agent.tools import get_all_tools
from src.core.cot_logger import log
JWT_SECRET = os.getenv("JWT_SECRET", "super-secret-key")
MAX_FILE_SIZE_BYTES = 10 * 1024 * 1024 # 10 MB
# In-memory per-user lock (keyed by user_id)
_user_locks: dict[str, bool] = {}
# In-memory service JWT cache
_service_jwt_cache: dict[str, str] = {} # {token: expiry_timestamp}
# #region AgentChat.GradioApp.Handler [C:4] [TYPE Function] [SEMANTICS agent-chat,handler,streaming]
# @ingroup AgentChat
# @BRIEF Core streaming handler — runs LangGraph agent, yields ChatMessage tokens with structured metadata.
# @PRE JWT valid, user authenticated.
# @POST Tokens streamed via yield; HITL interrupts yield confirm_required metadata.
# @SIDE_EFFECT Calls LLM, invokes tools, writes checkpoints.
# @RATIONALE Async generator pattern chosen for Gradio ChatInterface compatibility — Gradio iterates
# the generator and sends yielded JSON strings as event data to the frontend.
# @REJECTED Returning a single response (non-streaming) was rejected — violates FR-003 (streaming mandate).
async def agent_handler( # noqa: C901 — intentionally complex C4 orchestration
message,
history: list, # noqa: ARG001 — Gradio ChatInterface requires this parameter
request: gr.Request,
conversation_id: str | None = None,
action: str | None = None,
) -> AsyncGenerator[str]:
"""Handle incoming chat message. Streams tokens with structured metadata.
Args:
message: str or dict (when multimodal) — user message.
history: list of ChatMessage — Gradio's built-in history (ignored — loaded from DB).
request: gr.Request — may contain Authorization header with user JWT.
conversation_id: str — via additional_inputs (thread_id for checkpointer).
action: str — "confirm" | "deny" for HITL resume, None for normal messages.
"""
# ── Auth: extract user JWT if available —─
# Gradio runs behind Vite proxy which already handles auth.
# @gradio/client does not forward Authorization headers,
# so we don't enforce JWT here. Tool calls use SERVICE_JWT (see tools.py).
# The JWT is only used for user-scoped features (per-user lock, conversation context).
auth_header = request.headers.get("authorization", "")
user_jwt_str = ""
if auth_header.startswith("Bearer "):
try:
token = auth_header.split(" ")[1]
jwt.decode(token, JWT_SECRET, algorithms=["HS256"])
user_jwt_str = token
except jwt.InvalidTokenError:
pass # Ignore invalid JWTs — fall back to default context
# Store in ContextVar for @tool functions
set_user_jwt(user_jwt_str)
# ── Per-user lock (prevent concurrent sends per user) ──
user_id = _extract_user_id(user_jwt_str) if user_jwt_str else f"anon_{conversation_id or 'default'}"
if _user_locks.get(user_id, False):
yield json.dumps({"metadata": {"type": "error", "code": "CONCURRENT_SEND"}})
return
_user_locks[user_id] = True
try:
# ── Handle file upload ──
text = message.get("text", "") if isinstance(message, dict) else str(message)
files = message.get("files", []) if isinstance(message, dict) else []
if files:
# File size validation
file_path = files[0] if isinstance(files[0], str) else getattr(files[0], "name", None)
if file_path and os.path.exists(file_path):
file_size = os.path.getsize(file_path)
if file_size > MAX_FILE_SIZE_BYTES:
yield json.dumps({
"content": f"❌ File exceeds 10MB limit ({file_size / 1024 / 1024:.1f} MB)",
"metadata": {"type": "error", "code": "FILE_TOO_LARGE", "detail": "Max file size is 10 MB"},
})
return
parsed = parse_upload(files[0])
text = f"{text}\n\n--- Uploaded file content ---\n{parsed}"
# ── HITL resume path ──
if action in ("confirm", "deny"):
async for chunk in _handle_resume(conversation_id, action):
yield chunk
return
# ── Normal send path ──
conv_id = conversation_id or str(uuid.uuid4())
agent = create_agent(get_all_tools())
# Try up to 2 times: catch OutputParserException and retry with stricter prompt
max_attempts = 2
for attempt in range(max_attempts):
try:
async for event in agent.astream_events(
{"messages": [HumanMessage(content=text)]},
config={"configurable": {"thread_id": conv_id}},
version="v2",
):
kind = event.get("event")
# Audit logging for tool events
if kind in ("on_tool_start", "on_tool_end", "on_tool_error"):
await log_tool_event(event, conv_id)
if kind == "on_chat_model_stream":
chunk = event["data"]["chunk"]
if hasattr(chunk, "content") and chunk.content:
yield json.dumps({
"content": chunk.content,
"metadata": {"type": "stream_token", "token": chunk.content},
})
elif kind == "on_tool_start":
tool_name = event["name"]
yield json.dumps({
"content": f"🛠️ {tool_name}",
"metadata": {"type": "tool_start", "tool": tool_name, "input": event["data"].get("input", {})},
})
elif kind == "on_tool_end":
tool_name = event["name"]
output = event["data"].get("output", "")
yield json.dumps({
"content": f"{tool_name}",
"metadata": {"type": "tool_end", "tool": tool_name, "output": {"result": str(output)[:500]}},
})
elif kind == "on_tool_error":
tool_name = event["name"]
err = str(event["data"].get("error", "Unknown"))
yield json.dumps({
"content": f"{tool_name}{err}",
"metadata": {"type": "tool_error", "tool": tool_name, "error": err},
})
elif kind == "on_chain_end" and "interrupt" in event:
yield json.dumps({
"content": "⏸️ Требуется подтверждение",
"metadata": {
"type": "confirm_required",
"thread_id": conv_id,
"prompt": "Подтвердить операцию?",
},
})
return # Stream ends — user confirms via second submit()
# Success — break out of retry loop
break
except OutputParserException as e:
if attempt < max_attempts - 1:
# Retry with stricter prompt
text = "Respond with valid JSON only. Previous response was malformed.\n\n" + text
continue
# Final failure — yield error event
yield json.dumps({
"content": "❌ Ошибка обработки ответа LLM. Пожалуйста, уточните запрос.",
"metadata": {"type": "error", "code": "LLM_MALFORMED_OUTPUT", "detail": str(e)},
})
# ── Save conversation to DB via FastAPI REST ──
await _save_conversation(conv_id, text)
finally:
_user_locks[user_id] = False
# #endregion AgentChat.GradioApp.Handler
async def _handle_resume(conversation_id: str, action: str) -> AsyncGenerator[str]:
"""Resume from HITL checkpoint."""
agent = create_agent(get_all_tools())
if action == "confirm":
agent.invoke(
Command(resume={"action": "confirm"}),
config={"configurable": {"thread_id": conversation_id}},
)
yield json.dumps({
"content": "▶️ Операция подтверждена",
"metadata": {"type": "confirm_resolved", "result": "confirmed"},
})
elif action == "deny":
agent.invoke(
Command(resume={"action": "deny"}),
config={"configurable": {"thread_id": conversation_id}},
)
yield json.dumps({
"content": "⏹️ Операция отменена",
"metadata": {"type": "confirm_resolved", "result": "denied"},
})
def _extract_user_id(jwt_str: str) -> str:
try:
payload = jwt.decode(jwt_str, JWT_SECRET, algorithms=["HS256"])
return payload.get("sub", payload.get("user_id", "unknown"))
except Exception:
return "unknown"
# ── Conversation persistence ──────────────────────────────────────
SAVE_API_URL = os.getenv("FASTAPI_URL", "http://localhost:8000") + "/api/agent/conversations/save"
async def _save_conversation(conv_id: str, user_text: str) -> None:
"""Save conversation to DB via FastAPI REST.
Called after streaming completes. Creates or updates AgentConversation.
Uses SERVICE_JWT for auth. Failures are logged but not propagated.
"""
try:
service_token = os.getenv("SERVICE_JWT", "")
headers = {"Content-Type": "application/json"}
if service_token:
headers["Authorization"] = f"Bearer {service_token}"
async with httpx.AsyncClient(timeout=10) as client:
await client.post(
SAVE_API_URL,
json={
"conversation_id": conv_id,
"title": user_text.strip()[:100] or "Agent conversation",
"user_id": "0a82894e-d144-474b-aa61-81be2643d569",
},
headers=headers,
)
except Exception as e:
log("AgentChat.GradioApp", "EXPLORE", "Failed to save conversation",
{"conv_id": conv_id}, error=str(e))
# ── Gradio interface ──
def create_chat_interface():
"""Create the Gradio ChatInterface."""
return gr.ChatInterface(
fn=agent_handler,
type="messages",
multimodal=True,
additional_inputs=[
gr.Textbox(label="conversation_id", visible=False),
gr.Textbox(label="action", visible=False),
],
examples=[
["Покажи дашборды", None, None],
["Статус системы", None, None],
["Запусти миграцию", None, None],
],
)
# ── Healthcheck ──
async def health():
"""Healthcheck endpoint for Docker."""
return {"status": "ok", "uptime": os.times().elapsed if hasattr(os.times(), "elapsed") else 0}
if __name__ == "__main__":
demo = create_chat_interface()
demo.launch(
server_name=os.getenv("GRADIO_SERVER_NAME", "0.0.0.0"),
server_port=int(os.getenv("GRADIO_SERVER_PORT", "7860")),
)
# #endregion AgentChat.GradioApp