managed runtime
The runtime, operated.
Same runtime as the open-source Mercury — patched, scaled, and operated by us. Bring your own tools, your own model providers, your own code. We handle the plumbing.
cosmic stack cloud
The managed runtime for agents.
Run agent fleets in production without operating the runtime yourself. Cosmic Stack Cloud is the same open-source runtime that powers Mercury — managed, scaled, and observable, on infrastructure designed for long-horizon tool use.
Early access·GA target Q3 2026·200 early-access slots
statusEarly accessga targetQ3 2026runtimeOSS-compatiblepricingPer taskcomplianceSOC 2 · HIPAA
what cloud is
Six things you get the day you stop self-hosting.
long-horizon scheduling
Workloads that run for hours.
Agent jobs are not request/response. A single task can fan out across dozens of tool calls over minutes or hours. Cloud is built around that shape — checkpointed state, resumable on preemption, billed by useful work done, not wallclock.
observability built-in
Traces that explain themselves.
Every tool call, every model call, every retry, every cost — instrumented as a single trace per task. Search by failure mode. Replay any run. Diff two model versions on the same input. No third-party APM required.
byo provider
Your model contracts, your control.
Bring your OpenAI, Anthropic, Bedrock, Vertex, or self-hosted endpoints. We route, we fail over, we cache — we never proxy your tokens through our account or buy on your behalf.
compliance
SOC 2, HIPAA, data residency.
SOC 2 Type II audit underway. HIPAA BAA available for healthcare customers. Choose data residency per workspace: US, EU, or APAC. Audit logs exported to your SIEM.
pricing
Pay for useful work.
Per successful task, not per CPU-minute. Failed runs that didn't accomplish anything aren't billed. Pricing scales sublinearly with volume — large customers get genuinely lower unit costs, not 'enterprise pricing' theatre.
platform
Operated like infrastructure, not a SaaS.
| Regions | US · EU · APAC | Data residency per workspace |
|---|---|---|
| Uptime SLO | 99.95% | Tracked publicly at status.cosmicstack.ai |
| Compliance | SOC 2 · HIPAA | Type II audit in progress |
| Egress | Zero markup | Cloudflare-backed, pass-through pricing |
early access
Request access. We let in 5–15 teams per week.
Tell us about the workload. The more concrete you are about what you're building and what's currently in your way, the faster we'll get back to you.
faq
The questions we keep getting.
How is Cloud different from the open-source Mercury runtime?+
Same runtime — Cloud just operates it for you, with managed scaling, observability, multi-tenant isolation, and compliance. You can migrate either direction at any time without code changes.
Do I need to use your model providers?+
No. You bring your own provider credentials and we never proxy your tokens. We can also route to self-hosted models over private network.
What's the early-access process?+
Join the waitlist with a short description of your workload. We let in 5–15 teams per week, prioritized by fit (production agent workloads first, evaluation harnesses second, hobbyist projects when capacity allows). Expect 2–6 weeks.
When does Cloud become generally available?+
Target Q3 2026. We will not GA until the runtime has been operating customer workloads at scale for two full quarters.