Context
This epic owns the "works with the agent you already use" surface and the high-intent comparison queries. An /integrations hub with a per-tool copy-paste quickstart mirrors how E2B and Daytona convert, and a comparison cluster lets Mitos own the open-source / self-hosted / Kubernetes-native quadrant that no competitor fills.
The strategic spine (from the use-case and positioning research)
- Category narrative everyone now uses: "a computer per agent." The one validated way to explain fast forking is the
fork(2) system call applied to a whole computer: subprocess speed with full hypervisor isolation. Lead with it.
- Closest competitor is Daytona (fork + snapshot + per-agent-computer). Our wedge: microVM by default (Daytona's default is a container; isolation depth ranks Firecracker microVM > gVisor > container), live fork-to-many shipping today (Daytona's parallel-fork API is still partly roadmap), and OSS + self-host + Kubernetes-native (Modal cannot self-host and frames roll-your-own-k8s as a steep ops burden, which is exactly the burden we remove).
- Honesty rule: competitor latency figures are their published numbers on different hardware; ours stay reproducible from
bench/. Never describe a system that does not exist.
Sub-issues
Backed by engine work
Related existing work (do not duplicate)
Context
This epic owns the "works with the agent you already use" surface and the high-intent comparison queries. An
/integrationshub with a per-tool copy-paste quickstart mirrors how E2B and Daytona convert, and a comparison cluster lets Mitos own the open-source / self-hosted / Kubernetes-native quadrant that no competitor fills.The strategic spine (from the use-case and positioning research)
fork(2)system call applied to a whole computer: subprocess speed with full hypervisor isolation. Lead with it.bench/. Never describe a system that does not exist.Sub-issues
Backed by engine work
Related existing work (do not duplicate)