Free 3 min read

Is LangSmith Down?

Check live LangSmith status — LLM tracing, evaluation runs, datasets, prompt hub, and LangChain integration. See recent incidents and set up free email alerts.

LangSmith live status

LangSmith — live status

Updated every 5 minutes. Full incident history at prismix.dev/service/langsmith.

Full status →

Quick check: is LangSmith down right now?

  1. Prismix: prismix.dev/service/langsmith — live status + 30-day uptime + incidents.
  2. LangSmith status: langsmith.statuspage.io — LangChain's official status page for the LangSmith tracing and evaluation platform.
  3. API call: curl https://prismix.dev/api/v1/statuses | jq '.services[] | select(.id=="langsmith")'

Set up free email alerts for LangSmith

  1. 1

    Sign in

    Go to prismix.dev/sign-in — email OTP or GitHub sign-in.

  2. 2

    Star LangSmith

    On prismix.dev/service/langsmith, click the ☆ star icon.

  3. 3

    Alerts are live

    You'll get an email within minutes of any status change.

Common causes of "LangSmith not working"

If Prismix shows LangSmith as "Operational" but you're having issues:

  • Traces not appearing (missing LANGCHAIN_TRACING_V2 env var) — LangSmith tracing is opt-in and requires two environment variables: LANGCHAIN_TRACING_V2=true and LANGCHAIN_API_KEY=<your-key>. If either is missing or set after the LangChain import, traces are silently discarded. Verify both variables are set in the process environment before importing any LangChain modules.
  • Evaluation run stalled — LangSmith evaluation runs execute LLM-as-judge calls for each example in your dataset. If the evaluator model (often GPT-4 or Claude) is slow or rate-limited, the evaluation can pause for extended periods without an obvious error. Check the run's "Traces" tab in the LangSmith UI to see which evaluator calls are pending and which provider is the bottleneck.
  • Dataset upload too large — LangSmith datasets have a per-example and total-size limit. Very large datasets (millions of examples or examples with large embedded images/documents) can fail to upload via the UI or the SDK with a timeout error. Split large datasets into batches and use the SDK's client.create_examples() method in chunks of 100–500 examples.
  • API key missing required permissions — LangSmith API keys can be scoped to specific permissions (read-only, dataset-write, full-access). If you create a read-only key and then try to push traces or create datasets, operations silently fail with a 403. Generate a new key with the correct permissions in the LangSmith organization settings.
  • Hub pull rate limit exceeded — LangChain Hub (hub.pull("owner/prompt")) is rate-limited per API key. Applications that pull prompts on every request (instead of caching the pulled prompt object) quickly exhaust the limit and receive 429 errors. Cache pulled prompts at startup or use a functools.lru_cache decorator to avoid redundant hub calls.
  • Webhook not firing on run completion — LangSmith supports webhooks that notify an endpoint when a trace or evaluation run completes. Webhooks fail silently if your endpoint URL is not publicly accessible, returns a non-2xx status, or times out within 5 seconds. Test webhook delivery manually by triggering a test event from the LangSmith webhook settings page.
🔔

Stop manually checking — get alerts instead

Star LangSmith on Prismix and get emailed the moment status changes. Free, no credit card.

Monitor related LLM observability & AI tools?

Full status dashboard: prismix.dev/status