Operate · 6 min read
Deployment
Run LangChain applications in production with observability.
A LangChain application is ordinary server code. The operational work is mostly about timeouts, retries, caching and visibility.
Checklist
- Set explicit request timeouts on every model call.
- Retry on transient provider errors with exponential backoff.
- Cache embeddings; they are deterministic and expensive to recompute.
- Log the prompt version with every trace so you can attribute regressions.
- Put a hard spend limit on each API key.
Tracing
bash
export LANGCHAIN_TRACING_V2="true"
export LANGCHAIN_API_KEY="ls-..."With tracing enabled every run records its inputs, outputs, latency and token usage, which turns debugging a vague complaint into reading one trace.