Lumen Quickstart In development

Get it done in 10 minutes: install → integrate with LangGraph → view your first Trace → trigger a Replay → tally 24h costs.

10 minend to end

3 linesof integration code

0business-logic changes

Prerequisites

Python 3.9+ · Lurus API Key (how to get one) · an existing LangGraph/LangChain Agent project (or use the minimal example below if you don’t have one).

Install and integrate with LangGraph in three lines

bash

pip install lumen-ai

python

from lumen_ai import LumenTracer, LumenCheckpointer
graph = workflow.compile(
    checkpointer=LumenCheckpointer(),   # 崩溃恢复
    callbacks=[LumenTracer()],           # 执行追踪 + 成本追踪
)

No business-logic changes required — every LLM call, tool call, and state change is recorded to the Lumen backend.

Run a minimal example

python

from langgraph.graph import StateGraph, END
from lumen_ai import LumenTracer, LumenCheckpointer
from typing import TypedDict

class State(TypedDict):
    query: str
    result: str

def search(state: State) -> State:
    return {**state, "result": f"已处理: {state['query']}"}

wf = StateGraph(State)
wf.add_node("search", search)
wf.set_entry_point("search")
wf.add_edge("search", END)

graph = wf.compile(
    checkpointer=LumenCheckpointer(),
    callbacks=[LumenTracer()],
)

graph.invoke({"query": "什么是 WAL"})

View the Trace
Open the console (default http://localhost:7070) or the Lumen page in the Lurus console: the complete timeline of each graph run, the input/output/duration of every node, and the prompt/completion/token counts for each LLM call.
Trigger a Replay
Replay a past execution sequence once, without consuming tokens (useful for reproducing bugs locally, verifying fixes, and Prompt A/B testing):
python
```
from lumen_ai import Replay
Replay.from_run_id("run_abc123").play()
```
1
2

Tally 24-hour costs

python

from lumen_ai import CostTracker

print(CostTracker.summary(hours=24))
# { total_tokens: 128_340, cost_rmb: 12.47, by_model: {...} }

Replay costs nothing

Replay plays back from the recorded execution sequence and does not call the LLM again, so reproducing bugs, verifying fixes, and running Prompt A/B tests all cost zero tokens.

Next Steps

Python SDK — a deep dive into LumenTracer / LumenCheckpointer / CostTracker CLI Handbook — common commands and automation scaffolding Ecosystem Integration — working with Kova / LangGraph / OpenTelemetry

Lumen Quickstart In development ​

Next Steps ​

Next Steps

Related Products

Commonly paired with

Referenced by these products

Lumen Quickstart In development

Next Steps