// the find
tonbo-io/tonbo
Tonbo is an embedded database for serverless and edge runtimes.
Tonbo is an LSM-tree storage engine that writes Parquet SSTables to object storage (S3, R2, MinIO) and coordinates through a manifest with compare-and-swap commits — no long-running server required. It targets serverless and edge runtimes where you need durable structured state but can't run a database sidecar. Currently alpha, so the API is still moving.
The open-format bet is real: your data stays as standard Parquet files readable by DuckDB, Spark, or anything Arrow-native, so you're not locked into Tonbo's query path. The Arrow-native schema system with zero-copy RecordBatch reads is well thought through — projection pushdown means you don't pay for columns you don't touch. MVCC with snapshot isolation and time-travel queries is more than you'd expect from something billing itself as 'embedded'. The RFC-driven design docs (15 of them) show the team is thinking carefully about correctness, not just shipping features.
Filter pushdown is still in-progress — predicates are listed as a feature but the storage layer doesn't evaluate them yet, so you're pulling more data than you should be for selective queries. Multi-writer support is limited to single-writer-multiple-reader; concurrent writers would require CAS retries and the contention story at scale isn't documented. DataFusion integration is on the roadmap but absent, meaning you have to write your own query layer if you want SQL. Alpha status means the 0.4.0-a1 version tag isn't a formality — APIs have changed between releases and will again.