Supabase Compute runs agents and Dockerfile services beside the database in a private alpha. Not for production, and every service gets all project secrets.
Supabase launched Compute on Oct. 27, a private alpha that runs long-lived services such as AI agents, job processors and Dockerfile-based APIs inside the same project as the Postgres database. It targets work that outgrows Edge Functions. Supabase’s warning: “It’s for evaluation, not production, and it may change or end” (Supabase blog, Oct. 27, 2026).
What It Runs #
Supabase said some work doesn’t fit in an Edge Function: it needs more memory or CPU, runs longer than 400 seconds, or is written in another language. Compute gives each service a full Linux environment. You deploy Node, Deno or any Dockerfile, and pick 2 GB with 1 vCPU or 4 GB with 2 vCPUs. A service can be public, with a URL, or private, with none.
Deployment uses the Supabase CLI, v2.120.0 or later. You set compute = true under [experimental] in supabase/config.toml, create a service with supabase compute new and ship it with supabase compute push. Logs appear on a Compute page in the dashboard. The Management API has alpha endpoints to deploy, list and delete services.
Supabase cited Dreambase, which moved its data agent onto Compute. Its Postgres queries took a median of 2.9 ms, versus 65 ms from another region. The figure is Supabase’s, for one customer.
The limits #
Access is by waitlist. Only projects in us-west-2 qualify. Supabase listed the rest:
- A service can be restarted at any time. Save long-running work to Postgres.
- There is no autoscaling, GPU or persistent disk.
- Public services suspend when idle. Private services keep running.
- Sandboxes for AI agents aren’t part of the alpha.
- Every service gets all of the project’s secrets.
Supabase said waitlisted projects can try the alpha for free.
The take #
Teams that need a worker next to their data usually add a second platform with separate deploys and logs. Compute removes that.
The secrets rule is the one to read twice. A service running agent-written or third-party code with the project’s secrets in reach is a bigger risk than an Edge Function scoped to one job. Keep the alpha to throwaway projects, away from customer data.