Infrastructure / Deployment

Google Cloud Run

Fully managed serverless container platform with gVisor-based sandboxing, pay-per-request pricing, and scale-to-zero. Picked over Lambda for any-OCI-image flexibility and longer max execution. In 2026 Cloud Run added 'Cloud Run Instances' for long-running background agents and deep Gemini Enterprise Agent Platform integration, making it Google's go-to closed-source sandbox for agent code.

Google Cloud Run, fully-managed serverless container platform. Confirmed live June 2026 via cloud.google.com/run. At Next '26 Google announced a built-in `sandbox` tool for agents to run AI-generated code in ephemeral, strictly-isolated environments; container isolation historically backed by gVisor. Scored as the proprietary/SaaS deployment counterpart per the recipe (Cloud Run = closed counterpart).

Openness

1 high confidence

1.0

license: Proprietary
service: proprietary(Google SaaS)
source: closed
isolation-runtime: gVisor(open source, github.com/google/gvisor) but the Cloud Run product/control-plane is proprietary

Proprietary managed serverless platform; only the underlying gVisor sandbox runtime is OSS, not Cloud Run itself.

https://cloud.google.com/run recorded 2026-06-04
Cloud Run described as fully-managed Google serverless container product (proprietary SaaS)
https://github.com/google/gvisor recorded 2026-06-04
gVisor (Cloud Run's isolation runtime) is the open source component, distinct from the closed product

Adoption

5 high confidence

5.0

Google Next '26 blog (primary, verified verbatim): external active developers and apps on Cloud Run doubled YoY, more new customers/apps in 2025 than its first 6 years combined; Replit VP quote: >1M live projects hosted on Replit with Cloud Run as primary target. Hyperscaler serverless reach is well into the >10M band.

https://cloud.google.com/blog/products/serverless/whats-new-for-cloud-run-at-next26 recorded 2026-06-04
active developers/apps doubled YoY; more new customers in 2025 than first 6 years combined; Replit >1M live projects on Cloud Run

Capability

4 medium confidence

4.0

Very broad runtime/language coverage and hyperscaler scale; new agent `sandbox` tool (announced Next '26, 'coming soon') adds first-class AI-code-execution via a `sandbox do --` command. gVisor isolation is process/syscall-level rather than full microVM (Sprites/Northflank), so not the top isolation tier. Strong but not the single frontier-definer. Independent ComputeSDK TTI benchmarks corroborate the sub-second cold-start (~430ms median, 10th of 19 sandbox providers).

https://cloud.google.com/blog/products/serverless/whats-new-for-cloud-run-at-next26 recorded 2026-06-04
built-in `sandbox` tool for agents to run AI-generated code in ephemeral isolated envs; container model, autoscale 0->thousands
https://cloud.google.com/run recorded 2026-06-04
OCI container deploy, scale-to-zero, broad runtime support
https://www.computesdk.com/benchmarks/sandboxes/ recorded 2026-07-09
ComputeSDK TTI leaderboard (independent, 100 iters/day, 2026-07-09 run): median time-to-interactive ~430ms sequential / ~510ms burst @100 concurrent; 10th of 19 providers

Unchanged since 2026-07-30 (last edited, not re-checked)