from langgraph.graph import StateGraph
# Definējiet stāvokļa struktūru (piemēram, vaicājums, rezultāti, kopsavilkums)
class State(dict):
query: str
results: list
summary: str
builder = StateGraph(State)
@builder.node("search")
def search_node(state: State):
# izsauciet meklēšanas rīku
state["results"] = my_search(state["query"])
return state
@builder.node("summarize")
def summarize_node(state: State):
state["summary"] = summarize(state["results"])
return state
builder.edge("search", "summarize", condition=lambda s: len(s["results"]) > 0)
app = builder.compile

from langgraph.graph import StateGraph
from typing import List, Optional
class State(dict):
query: str
plan: List[str]
step: int
artifacts: List[str]
draft: str
confidence: float
builder = StateGraph(State)
@builder.node("plan")
def plan_node(state: State):
state["plan"] = make_plan(state["query"]) # LLM-structured list
state["step"] = 0
state["artifacts"] = []
return state
@builder.node("act")
def act_node(state: State):
task = state["plan"][state["step"]]
output = execute_task(task) # tool(s)
state["artifacts"].append(output)
return state
@builder.node("synthesize")
def synth_node(state: State):
state["draft"] = synthesize(state["artifacts"]) # LLM combine
return state
@builder.node("evaluate")
def eval_node(state: State):
score, feedback = evaluate(state["draft"]) # rubric-based
state["confidence"] = score
state["feedback"] = feedback
return state
# Saites
builder.edge("plan", "act")
def more_steps(s: State) -> bool:
return s["step"] < len(s["plan"]) - 1
builder.edge("act", "act", condition=lambda s: (s.update({"step": s["step"] + 1}) or True) and more_steps(s))
builder.edge("act", "synthesize", condition=lambda s: not more_steps(s))
builder.edge("synthesize", "evaluate")
builder.edge("evaluate", "plan", condition=lambda s: s["confidence"] < 0.7) # plāna uzlabošana
app = builder.compile