Introducing Supabase Compute, in private alpha. You don’t need a second platform for your agents and long-running jobs. Run them in your Supabase project, next to your data in Postgres. See how it works →

Compute is for work that doesn’t fit in an Edge Function, like an agent that runs for an hour or an API written in Python. Your services get your project’s secrets, and their logs show up in your dashboard. Join the waitlist to try it for free →

Check out what people have been building with Compute 💚

This is amazing. You’ll find this really helpful if you’re a real developer. If you are just pretending and messing around, you don’t even get it: auto selection at this point.

junior developers when they get to the word "jobs":

Nice job guys.

supabase slowly becoming the whole cloud 😭

my agents already talk to supabase on feednest, thats how i fixed a lot of slow queries. running them right next to postgres, ill definitely try this

goodbye obscure serverless timeout limits, hello 1-hour agent runs

Running agents right next to Postgres is nice until one of them holds a transaction open for 40 minutes. Are there per-job connection limits, or does that come out of the project's pool?

This is Supabase coming for the background-jobs market Render, Railway and own. Agents next to the data cut network hops and auth headaches, but the real test is pricing: long-running agents burn compute, and alpha pricing rarely survives launch.

Congrats on the alpha! Does Supabase Compute support UDP? Thinking QUIC / HTTP/3 for real-time workloads.

a second platform was also the exit. jobs on Render can point at any Postgres, jobs inside the Supabase project only move if the whole project moves. good to know before the agent code grows around it.

The line I'd read twice is "every service gets all of your project's secrets". Fine with one agent, but once I've got a few agents sitting next to prod data I want secrets scoped per service. Is that coming after the alpha?

postgres slowly absorbing the entire stack, one private alpha at a time

Compute next to Postgres is the right call for agents: most agent latency is chatty DB round trips, not model time. What I'd want answered before moving jobs there is how it's isolated from the database's own CPU and memory, and what happens to in-flight jobs on a project restart.