TL;DR: LangSmith remains the reference for tracing and observability of LangChain agents. Its SaaS model, however, raises a sovereignty challenge. The open source Idun platform (GPL v3, public and auditable code) integrates LangSmith natively and lets you switch to Langfuse, Arize Phoenix or OpenTelemetry by changing 3 environment variables, without rewriting your LangGraph agents.
LangSmith remains the reference for tracing and LangSmith observability of LangChain agents in production. But as your agents gain autonomy, one question becomes unavoidable: does your tracing data stay under your control? The open source Idun platform integrates with LangSmith natively, from tracing to LangSmith evaluation, from LangSmith deployment to sovereign governance. Book a 30-minute discovery call.
LangSmith is an LLM observability platform built for LLM applications and AI agents. Its three core functions: tracing (recording every model call, with latency and cost), debugging (pinpointing failures and agent loops) and evaluation (measuring output quality with LLM-as-judge or custom evaluators).
According to LangChain's "State of Agent Engineering" report (2026), 89% of organisations running agents in production have some form of LangSmith observability or an equivalent. The trace is the reference diagnostic signal: without it, trying to debug agent failures means debugging blind. The LangSmith platform answers that need: LangSmith debugging, LangSmith evaluation and the ability to monitor quality are its three operational pillars. A client SDK is available for Python, JavaScript and any OpenTelemetry-compatible framework. Every agent run is recorded. Evaluation results are exported to the backend of your choice.
LangSmith is optimised for the LangChain ecosystem. The concrete limits for enterprise teams:
- Self-hosting: only available on the Enterprise plan (Kubernetes required). No simple Docker option for teams that want to host on their own infrastructure. - Vendor dependency: until March 2025, LangSmith did not support OpenTelemetry end to end. Exporting traces to an alternative backend required manual integration. - No runtime: LangSmith observes agents but does not run them. It provides no guardrails, no persistent memory and no access control at runtime level.
These limits do not disqualify LangSmith. They define what you need to add around it to reach production at enterprise scale.
The Developer plan offers 5,000 traces per month. The Plus plan is billed per seat per month, with additional traces billed on usage. For a team of 10, the monthly bill multiplies accordingly, before any overage charges.
By comparison, Langfuse offers 50,000 events per month on its entry tier (10 times more), with a self-hosted option and no licence fees. For clients who quickly outgrow LangSmith's entry tier, the cost gap becomes structural, not cyclical.
Connecting LangSmith to the open source Idun platform takes less than 5 minutes. Switching observability backend takes 30 seconds of configuration: replace the LangSmith environment variables with those for Langfuse or Arize Phoenix.
Step 1: Install
``bash pip install idun-agent-engine ``
Step 2: Configure LangSmith (environment variables)
``bash # .env - switch provider without touching the agent code LANGCHAIN_TRACING_V2="true" LANGSMITH_API_KEY="ls__votre_cle_api" LANGSMITH_PROJECT="mon-agent-production" ``
