💻 OpenSmith is an open-source tool for local tracing of LLM pipelines, serving as a privacy-first alternative to LangSmith.
The project allows users to track model execution steps, log data, and analyze costs using only local SQLite or Postgres. It supports Python decorators (@trace) and automatic patching for popular backends such as OpenAI, Anthropic, and LiteLLM.
🌍 The emergence of local-first observability tools reduces the risk of data leakage when debugging RAG systems and LLM agents, allowing for work without compliance violations.
👤 For AI application developers, this provides the ability to visualize the entire model workflow on their own machine while saving on cloud service subscriptions.