langprobe
Getting Started

Migrate from LangSmith

Point your existing LangSmith client at langprobe. One env var, no code change.

langprobe's ingest is LangSmith-compatible: it accepts the native RunCreate / RunUpdate shapes a LangSmith client already sends. Migrating is import-and-go — you don't touch your instrumentation code, you re-point it.

Re-point your client

Set the two environment variables your LangSmith SDK already reads to your langprobe host and key:

export LANGSMITH_ENDPOINT=http://localhost:7080
export LANGSMITH_API_KEY="lt_<public_id>.<secret>"

That's the whole change. Your existing traceable functions and callbacks keep running; the langprobe shim translates each RunCreate / RunUpdate into the same ingest queue that native runs land in, so they show up alongside everything else in /runs.

Swap http://localhost:7080 for your own ingest host (or the hosted https://app.langprobe.com). Mint the lt_… key in API keys — see the Quickstart.

What carries over

  • Runs and spans — the run tree, inputs/outputs, latency, and token/cost attributes map onto langprobe's run and span model.
  • Nothing to re-instrument — no new decorators, no new SDK. If it worked against LangSmith, it works here.

Once traces are landing, the parts LangSmith doesn't do are what you came for:

  • Replay — edit a prompt, model, or tool config, re-run, and diff span by span with a determinism verdict.
  • Agent surface / MCP — token-budgeted, LLM-legible run views so an agent can debug an agent.
  • Eval-rigor — judge your judges before you trust a score.

On this page