SmithDB is a distributed database designed for agent observability, serving as the data layer for LangSmith. It processes trace data from AI agents, handling deeply nested spans and asynchronous event arrivals. The architecture relies on object storage with stateless ingestion and query services to support deployment in self-hosted and multi-cloud environments.[02]
- Country of Origin
- US
- Start Year
- 2026
- Project Type
- Internal / Non-Public
- Written in
- Rust
- License
- Proprietary
SmithDB is a distributed database designed for agent observability, serving as the data layer for LangSmith. It processes trace data from AI agents, handling deeply nested spans and asynchronous event arrivals. The architecture relies on object storage with stateless ingestion and query services to support deployment in self-hosted and multi-cloud environments.[02]
History
Announced by LangChain in May 2026, SmithDB was developed to address the scaling limitations of traditional observability stores when handling modern AI agent traces. It now backs core LangSmith workloads, replacing previous storage solutions to accommodate increased trace volume, payload size, and complex query patterns.
Citations
2 sources- We built SmithDB, the data layer for agent observability langchain.com
- SmithDB • Buttondown buttondown.com