Platform and availability

Current compute, runtime, and access information. For dated component observations and incidents, see service status.

Compute placement and regions

Containers run on Cloudflare infrastructure. Mainbrella does not offer a customer-selected compute region or guaranteed data residency. The API does not publish the region of an individual container; Cloudflare’s network footprint is not a list of guaranteed Mainbrella regions.

Machines and runtimes

CapabilityCurrent availability
Machine sizeAll plans: five sizes, from Lite (1/16 vCPU, 256 MiB RAM, 2 GB disk) to XL (4 vCPU, 12 GiB RAM, 20 GB disk).
Default runtimeNode 24 image with shell access.
Other runtime imagesCatalog IDs may include Python, Rust, Go, and DevOps. Availability depends on the deployed catalog returned by GET /containers.
Custom imagesAccount-owned images can be launched when ready. Building depends on buildsEnabled from GET /images.
AccessHTTP execution and files, SSH through cloudflared, and a browser terminal.
Network and storageOutbound internet is enabled by default. Snapshot stop/resume, filesystem forks, destination rules, and secret injection require their corresponding flags from GET /capabilities. Saved workspaces can be restored into fresh containers. No memory/process resume or persistent volumes.

See Images for environment selection and Limits for command, file, connection, and build bounds.

Developer capabilities

Source implementation and deployed availability are separate. Check live capabilities before use. The following additions need matching API/runtime support and an SDK release.

FeatureStatus and limits
Stateful PythonSource implementation. Generation-scoped contexts; requires Python in the image. Contexts end on timeout, runtime restart or stop.
Streaming transfersSource implementation. Atomic file uploads and streamed downloads up to 64 MiB; separate from the basic file API.
PrewarmingSource server-managed pools with automatic replenishment, expiry and retry-safe acquisition; gated by pools.managed. SDK-local pools also remain available. Waiting inventory consumes normal start and compute budgets.
Image versions and tagsImmutable image IDs/digests with owner-scoped movable tags. Customer builds still require deployment configuration and qualification.
Image reuseSource owner-scoped build cache and digest-pinned private base imports. Requires a private build repository, migration, configuration and live qualification.
Desktop automationSource SDK helpers for Linux mouse/keyboard control, application launch and screenshots on compatible images. Uses bounded interpreter sessions. No VNC viewer, recording or Windows support.
Agent integrationsSource stdio MCP server plus published HTTP/SDK and skill integrations. No managed language server.
TelemetrySource OTLP JSON pull export for lifecycle logs and enabled workload metrics. Metrics and webhook release gates still apply.
Memory-preserving pauseUnavailable. Filesystem snapshots do not preserve RAM or running processes.
Managed persistent volumesUnavailable. External object-storage mounts require a compatible image, bucket credentials and runtime qualification; they do not provide local-disk semantics.

See the generated API and SDK references for usage and limits.

Internal networking

Private Services is a local HTTP prototype for connecting independently managed machines through named .internal endpoints. Explicit network membership and exact-generation checks govern access; a backend needs no public preview. Production support has not been qualified. Require networking.privateServices: true from GET /capabilities. Only bounded plain HTTP is supported, with no arbitrary TCP, HTTPS, WebSockets, private IPs or direct database protocols.

Plan capacity

PlanBillingUnit-hours/monthConcurrent units / containersStarts/UTC monthSession / idle
Prepaid$5 minimum top-up$0.02/unit-hour128 / 10010,00024 hours / 30 minutes
Builder (legacy)$525028 / 51,0001 hour / 10 minutes
Pro (legacy)$1809,000128 / 10010,00024 hours / 30 minutes
Scale (legacy)$99950,000640 / 500100,00072 hours / 60 minutes

Failed starts count toward the monthly start limit. Prepaid accounts consume purchased credit for elapsed compute, including startup and idle time, within their monthly spending cap; existing legacy plans retain fixed compute allowances. Concurrency and session limits also apply. Every size is available on every plan. Returned account limits govern access; check them before launching parallel work.

Service access and commitments

Sign in and add prepaid compute credit. Use the dashboard for interactive containers or the Quickstart for API access. Self-service accounts have no uptime SLA or guaranteed support response time.

Service status · Startup benchmarks · Trust center