OpenSmith – Local-First LLM Tracing

LangSmith requires cloud accounts just to debug your own LLM traces, adding friction and privacy concerns. OpenSmith is a self-hosted alternative using SQLite that runs locally with zero config. Target users are ML engineers and developers building with LLMs.

LangChain's LangSmith is the obvious incumbent here, and its mandatory cloud dependency has generated consistent complaints on HN and Reddit from developers working in air-gapped environments, handling regulated data, or simply unwilling to route traces through a third-party service — that friction is real and documented. The broader trend of local-first developer tooling (Ollama, LocalAI, etc.) has normalized the expectation that the entire LLM dev stack should run offline, which makes the timing credible. At $1k–5k/mo the revenue ceiling is low, and it likely stays that way unless there's a clear paid tier beyond the open-source core — individual devs won't pay much for a debug tool, so the path to meaningful revenue probably requires enterprise self-hosted licensing or support contracts, which is a harder sell than it looks. The biggest risk is that LangSmith ships an optional local mode or LangFuse (already open-source and self-hostable) captures this audience first, leaving no differentiated wedge.