Lumen Ecosystem Integration In development
As a semantic-layer observability tool, Lumen connects downward to the Kova execution engine and the Lurus API gateway, sits alongside LangGraph for compatibility, and exports standard OTel traces outward to Grafana / Datadog.
WAL persistence layer with microsecond-level crash recovery.
One-line Saver replacement.
Automatic token-billing correlation.
Export standard traces.
With Kova
Kova handles low-level WAL state persistence and microsecond-level crash recovery, while Lumen handles semantic-layer trace annotation and cost accounting.
from lumen_ai import LumenCheckpointer
from kova import KovaClient
kova = KovaClient("kova://prod")
graph.compile(checkpointer=LumenCheckpointer(kova_client=kova))With LangGraph
Drop-in Saver replacement (the API is fully compatible with BaseCheckpointSaver):
- from langgraph.checkpoint.sqlite import SqliteSaver
+ from lumen_ai import LumenCheckpointer
- graph = wf.compile(checkpointer=SqliteSaver.from_conn_string(":memory:"))
+ graph = wf.compile(checkpointer=LumenCheckpointer())With Lurus API
When calling an LLM through the Lurus API, token billing and latency are automatically correlated to the Lumen trace:
from openai import OpenAI
from lumen_ai import LumenTracer
client = OpenAI(base_url="https://hub.lurus.cn/v1", api_key="sk-...")
tracer = LumenTracer()
with tracer.span("classify"):
resp = client.chat.completions.create(...)
# 调用自动记录，鹿贝消费可追溯OpenTelemetry
Export as an OTel trace: LumenTracer(exporters=[to_otlp("http://otel-collector:4317")]) (from lumen_ai.otel import to_otlp).
Grafana / Datadog
The pipeline is as follows: the Lumen SDK pushes traces to the OTel Collector over the OTLP protocol, which then fans them out to Grafana Tempo or Datadog APM:
graph LR
A[Lumen SDK] -->|OTLP| B[OTel Collector]
B --> C[Grafana Tempo]
B --> D[Datadog APM]For Grafana dashboard templates, see the Lurus official template library (search for "Lumen Dashboard").