Computefor agentic workloads
Agent sandboxes and always-on backend services running inside your Supabase project — next to your data, behind your auth.
One runtime,two shapes of work
Sandboxes, APIs, and background jobs usually mean three vendors and a database somewhere else. Compute runs them all in Supabase alongside your application backend.
Isolated environments for agent code
Execute code in a secure, isolated environment — one per task, per session, or per agent.
- Per-session environment for executing code
- Provision programmatically via the API or MCP server
- Suspend and resume with state preserved
Always-on backends in any language
Transcription pipelines, embedding jobs, agent frameworks, and APIs — running as long as the work takes.
- Auto-scaling for traffic surges
- No wall-clock limits on execution
- Any runtime, static binary, or Dockerfile
- HTTP services, background jobs, and long-running pipelines
One primitive, every workload
Ephemeral sandboxes and always-on HTTP services, one primitive.
Any language
Node, Deno, or any Dockerfile.
Next to your database
Same network as Postgres — single-digit millisecond queries.
Scale to zero
Services suspend when idle and resume in under a second. You pay nothing while they sleep.
One trust boundary,preconfigured
Every sandbox and service inherits your project's auth, roles, and permissions the moment it starts, running in its own microVM with dedicated CPU and memory. Agents get a sandbox that is already scoped to the data it is allowed to touch — nothing more.
Short-lived credentials
Access runs through Supabase Auth with short-lived keys, not standing secrets.
Zero-config data access
Reach your database and Storage with the permissions your key already carries.
Per-workload firewalls
Define which external endpoints and ports each workload can reach.
Cloaked secrets
Scope secrets per workload, and cloak values so your code never sees them.
Patched automatically
Kernel-level security patches roll out automatically. Nothing to do on your part.
Dependency scanning
Dependencies are scanned at runtime, with alerts as new vulnerabilities land.
Deploy fromwherever you work
The same deployment surface serves humans and agents: one command in a terminal, one tool call in an agent session, one request against the API.
Join the waitlistFrequently asked questions
Compute complements Edge Functions, which remain fully supported and best for short-lived work like webhooks and lightweight APIs. Compute handles what they can't: long-running jobs, background workers, and heavier workloads with no wall-clock limits. If a function outgrows Edge Functions, you can migrate it to Compute without code changes.
In Private Alpha, Node, Deno, and any Dockerfile are supported — no wrapper scripts or extra orchestration code required. Bun and Python are on the way; until they land, a Dockerfile covers anything else you need to run.
There are no wall-clock limits, so jobs and pipelines run as long as the work takes. For HTTP services, Compute suspends automatically when idle on network I/O and resumes in under a second, so a service configured as always-on behaves like one without billing you for idle time.
During Private Alpha, Compute is only available in us-west-2, with more regions rolling out over time until every Supabase region is covered. Sandboxes and services deploy to your primary database's region — that's where writes happen, and it keeps behavior predictable. They can still read from replicas via the load-balanced connection endpoint, so replica capacity isn't wasted. Pinning a workload to a specific replica region is on the roadmap.
Sandboxes and services are designed to scale to zero and restart on demand, so the local filesystem is fast scratch space, not durable storage. For state that needs to persist, use your Postgres database for structured data and Supabase Storage for files — both reachable with near-zero intra-region latency and no data-transfer tax. Mountable persistent disks are on the roadmap.
Compute bills per compute-hour based on the instance size you choose, with no per-request metering. Idle sandboxes and services suspend automatically, so you pay nothing while they sleep. Traffic between a workload and its own database is free. There is no separate subscription — usage appears on your existing Supabase invoice with the same spend caps and credits. Final rates will be published before general availability.
Compute is in Private Alpha. Join the waitlist and we will send invites in waves as capacity grows. Feedback from alpha participants directly shapes the product ahead of public launch.
Run your next workloadnext to your data
Compute is in Private Alpha. Join the waitlist and we'll reach out as invites roll out.