ComputeSDK Actions replaces the execution layer of GitHub Actions. Keep your workflows, run them on the providers and regions you choose, and give your agents the whole loop, from dispatch to logs to rerun.
Connect a Git remote and run workflows by ref or commit SHA, without rewriting them for a provider-specific runner.
Live step logs and upload-artifact / download-artifact support, with artifacts kept for 30 days.
Scheduled workflows run on your cron, and keep working when GitHub Actions is down.
Set an ordered provider and region policy. Actions places each job with the first provider that accepts it and records every attempt.
Building a sandbox platform? Join the Actions provider network. Reach out in Slack, on X, or email support@computesdk.com.
Placement order
Reorder providers, pin regions where needed, and keep the same workflow as capacity moves.


Provider and region are inputs, not constraints inherited from your CI. Actions walks your ordered placement policy and runs the job on the first provider that accepts it. If one rejects the job or has no capacity, the next one takes it.
Git is the only hard dependency. Give Actions a remote, credentials, and a ref, and it discovers and runs the workflows in that repository. GitHub, GitLab, Bitbucket, or a self-hosted remote over HTTPS or SSH.
Actions runs supported GitHub Actions workflows as they are: existing actions, steps, secrets, matrices, schedules, and artifacts. Replacing the execution layer shouldn't mean rewriting your YAML.
Dispatch returns the run ID immediately, so the agent that started the work can follow that exact run: stream logs, find the failed step, pull artifacts, cancel, or rerun, all without a human with a browser.
Switch providers by changing the placement policy, not the workflow. Every placement attempt is recorded, so where a job ran is part of its output.
placement policy: namespace → tensorlake → blaxel → archil:sfo1
namespace placed sandbox i45vss5e0og5g
tensorlake placed sandbox 9kfpay6hco5geyhr130o7
blaxel placed sandbox 3659a0e6-04cd-44b0-a445-7bf30553e17a
archil placed sandbox 01a0e8d1-1b11-7a87-b87e-ba9bd8a1a940
job "test (node 22)" running on namespace, tensorlake, blaxel, archilStructured JSON output, stable error codes, scoped non-interactive auth, and log streaming, in the compute CLI and the v1 API.
compute actions dispatch acme/api --workflow ci.yml --ref main --json
# → { "runId": "run_8f2c…", "created": true, "headSha": "7fd1a60…", "url": "…" }
compute actions logs run_8f2c… --follow
compute actions summary run_8f2c… --json
compute actions rerun run_8f2c…Getting started
Install the GitHub App, or add any Git remote with a token or SSH key.
Add provider keys and set a placement order. Each key is tested by booting a real sandbox before a job ever uses it.
Workflows run on push, on schedule, or on demand from the dashboard, CLI, or API. Status lands back on the commit.
Pricing
Full details on the pricing page.
Agents are making development faster and more parallel. The verification loop shouldn't be an opaque queue or a browser-only debugging session.
FAQ
For the execution layer, yes. You keep your workflow files and your Git host; Actions takes over scheduling, placement, execution, and logs. You can move one workflow at a time.
No. Actions reads supported GitHub Actions YAML as-is. Anything outside the supported surface is refused before a sandbox boots, with the reason recorded on the run.
Today: Namespace, Tensorlake, Archil, Blaxel. We're building a qualified provider catalog from real compatibility, reliability, performance, and cost data, and adding providers as they pass.
You do, on your own provider accounts. The Actions subscription covers orchestration, dashboards, schedules, and the API.