Placement
How Boxport picks a provider for each sandbox.
Every create is placed by score. Lower wins. The score adds three terms, each scaled against the worst provider in the running so the weights compare like with like:
| Term | From | Default weight |
|---|---|---|
| Cost | List price for your size and shape | 3 |
| Latency | p95 create time over the provider's last 200 creates | 1 |
| Reliability | Create failure rate over the same window | 2 |
Health is pooled across every Boxport account and hourly probes, so a new account gets good placement from its first create. Pooled records carry only provider, outcome and time.
Shape matters
Tell Boxport how the session will run with shape: minutes it will last and the share of time the CPU is busy. Providers meter differently. E2B and Daytona bill every second the sandbox exists; Modal bills CPU only while it's used. So a long, mostly idle session is cheaper on Modal and a short busy one is cheaper on E2B or Daytona. Boxport does that sum for you.
Policy
"policy": {
"weights": { "cost": 1, "latency": 0, "reliability": 1 },
"allow": ["e2b", "modal"],
"maxSessionCost": 0.10,
"split": { "modal": 0.1 }
}
allow: only these providers.maxSessionCost: skip providers whose estimate is above this many dollars.split: send this share of creates to a provider first. Use it to move traffic to a cheaper provider gradually; a failed create there still falls back.
Failover
If a create fails, Boxport tries the next provider in order and returns every attempt. A provider with three failures in a row is skipped for 30 seconds. A wrong vendor key counts as auth: it falls back but never counts against the provider's health.
Dry run
POST /v1/place takes the same body as a create and returns the order, the score of each provider and why any were excluded. It creates nothing.