Availability & Roadmap
This page gives customers a clean read on what is ready for production today, what is available for selected partners, and what is still on the roadmap.
Release stages
| Stage | Meaning |
|---|---|
| Generally available | Supported for production use with documented APIs and normal support paths. |
| Partner preview | Available to approved customers while we finish polish, SDK parity, or operating runbooks. |
| In hardening | Built or partially available, but not ready to promise as a stable public surface. |
| Planned | On the roadmap, not available for production use yet. |
Generally available
Create isolated Linux workspaces for code generation, build steps, test runs, long-running commands, file operations, and live previews.
Expose a running process on a public preview URL, with private preview tokens for embedded product experiences.
Turn a sandbox into a durable deployment version, promote it, list versions, and roll back when needed.
Use Postgres, MySQL, Redis, Qdrant, object storage, per-project auth, and persistent volumes with credentials injected into runtime.
Use external IDs, browser tokens, custom branding, preview domains, webhooks, and usage rollups to build MIOSA into your own product.
Partner preview
These surfaces are being used with selected customers while we continue polishing public docs and SDK parity.
| Surface | Current scope |
|---|---|
| Computers | Full Linux desktops for browser automation, screenshot loops, and computer-use agents. |
| Dynamic deployments | Production server processes created from a sandbox publish. |
| Docker Host | Workspace-scoped container runtime for apps that need Docker, server processes, background workers, or database-backed production hosting. |
| Tenant deployment domains | Branded app domains for generated apps, separate from sandbox preview domains. Routing exists; tenant-level self-service remains operator-assisted while exact deployment domains are self-service. |
| External computer providers | Bring third-party computer providers into the same MIOSA computer abstraction. |
| Evaluations | Desktop and agent workflow benchmarks using reproducible tasks, traces, and scoring. |
In hardening
| Area | Status |
|---|---|
| Desktop computer provisioning | Computers preserve disks and snapshots across stop, resume, and failed recovery. Start latency and stream readiness continue to be hardened. |
| Dynamic runtime routing | Dynamic runtimes keep the active healthy version serving while candidates build. Route evidence and automated recovery continue to be hardened. |
| Docker Host fleet maintenance | Host upgrades preserve deployment records and desired state, restore desired-running apps, and verify workloads before returning a host to active service. |
| Customer-cloud regions | AWS and Google Cloud foundations exist, but production placement remains gated on a real provider canary and per-host acceptance evidence. |
| Managed data recovery | Database IDs, credentials, volumes, bindings, and desired state survive host maintenance. Recovery visibility continues to be expanded across product surfaces. |
| Deployment proof | CLI, TypeScript SDK, Python SDK, and Go SDK expose proof checks. Java and Elixir parity is still in progress. |
| SDK parity | Python and TypeScript are deepest today. Go, Java, Elixir, and UI packages are being brought to the same coverage over time. |
| Cross-workspace analytics | Resource attribution is live. Higher-level dashboards and billing views are expanding. |
What is live in Release 4
Recent platform changes customers can rely on now. Each is a result, not a promise of future work.
| Change | Result |
|---|---|
| Disk corruption guard | Restoring a memory snapshot onto a disk that grew after the snapshot is refused instead of corrupting the filesystem. Live on the control plane and all hosts. |
| Guest isolation from the internal network | A sandbox cannot reach the platform’s own network except the managed-database proxy range. The Firecracker rule ordering that stops an older allow rule from shadowing the deny is live fleet-wide, and SOMA already enforced it. |
| Preview fallback for white-label staff | Preview access for a white-label organization’s own staff falls back to the authenticated console path instead of failing. |
| Disk larger than the pre-built slot | A request for a disk larger than the pre-built slot boots at the requested size instead of failing or returning the wrong size. |
| Learn credits cooldown | An expired Learn attempt is timed from its expiry, and an expired attempt with no answers no longer counts as a failure. |
| Function pool | The per-host function pool is deployed but dark: it is not a customer surface yet. |
| SOMA catalog accuracy | The runtime catalog no longer advertises pause or snapshots for SOMA, which has neither. |
Planned
| Capability | Notes |
|---|---|
| Native GPU sandboxes | Targeted early 2027, for workloads that need accelerated inference or rendering. |
| Edge functions | Targeted Q1 2027, as a deployment target separate from origin runtimes. |
| Native nameservers | Targeted Q4 2026, for customers who want MIOSA to manage more of the domain lifecycle. |
| OTLP export | Planned for teams that want platform traces in their own observability stack. |
| Test-mode API keys | Planned to separate production billing from integration testing. |
Deliberate product boundaries
MIOSA is focused on AI-generated software, isolated execution, live previews, desktop automation, white-label SaaS infrastructure, and durable deployment. It is deliberately not:
- Managed Kubernetes. Sandboxes and Deployments cover the workloads people would otherwise use K8s for.
- Generic VPS hosting. Compute is always tied to a tenant identity (workspace or
external_workspace_id). - Cryptocurrency or blockchain primitives.
See also
- Benchmarks for verified performance results.
- Evaluations for agent task scoring and desktop workflow benchmarks.
- Provider comparison for the compute-provider landscape.
- Customer infrastructure for MIOSA-managed, connected-machine, AWS, Google Cloud, and hybrid placement.
- Known limitations for the concrete gaps and gotchas behind these stages.
- Changelog for shipped product changes.