Supabase Compute lets you run agents next to your database
A new private alpha runs long-lived services, background jobs and AI agents inside your Supabase project instead of on a separate platform.
Supabase launched Compute in private alpha on 27 October: a way to run long-lived services, background jobs and AI agents directly inside a Supabase project, next to the Postgres database, rather than on a separate platform.
It exists for the work that doesn’t fit an Edge Function: anything needing more memory or CPU, running longer than the 400-second Edge Function limit, or written in a language Edge Functions don’t support. Compute runs Node, Deno or a custom Dockerfile, with sizes of 2GB RAM and 1 vCPU or 4GB RAM and 2 vCPUs, for minutes or hours rather than seconds. You deploy from the Supabase CLI (2.120.0 or later) by adding compute = true under [experimental] in supabase/config.toml, then supabase compute new and supabase compute push.
[experimental]
compute = trueThe alpha is evaluation only: it runs in us-west-2 alone, with no autoscaling, GPU or persistent disk, and services can restart at any time. Public services suspend when idle; private ones keep running.
Why it matters: running an agent or a background job has usually meant standing up a separate service on something like Fly.io or a Vercel background function, then wiring it back to the database. Compute keeps that workload in the same project, with the same config and the same Postgres instance it’s reading from.
The caveat: it’s private alpha, one region, and not meant for production yet. Access is by waitlist.