Enterprises must move data between many sources and destinations — but a lot of that data lives on-prem or at the edge, behind firewalls, where you can't simply pull it into the cloud.
A control-plane / data-plane split. One streaming ETL engine is written once and runs identically in the cloud AND on a tiny, outbound-only edge agent — so a migration can execute right next to the data.
Outbound-only edge agent
No inbound ports; a long-poll job protocol. Edge secrets are resolved on-host and never sent to the cloud. Per-connection cloud/agent placement.
Atomic job leasing
Postgres FOR UPDATE SKIP LOCKED, with a heartbeat + lease-TTL reaper and exponential backoff so jobs never double-run or get stuck.
Durable runs
A River queue makes runs resumable and cancellable, with SSE live logs.
Pluggable blob seam
Filesystem or S3, with automatic direct-vs-staged execution, cursor-based incremental watermarks, and cron scheduling.
- Deployed live on Vercel, Render, Neon and Supabase S3.
- Validated by moving 1.72M+ rows — and growing.
- One Go monorepo: apiserver, worker, engine, agent.