Host capacity

What it is

Every host knows exactly what it is running and admits only what fits.

flashhostd sees every FlashVM on a machine: plain, pooled or sleeping. Each create passes a fit check against real use and predicted need. Events stream live, and metering records what every VM used, hour by hour.

Why it matters

Dense AI sandboxing runs on trust in the numbers. FlashVMM packs hosts tight because it knows the room before a VM starts, and bills each VM for exactly what it used.

• Admission first. Room is reserved before a VM boots, so a full host says so in one clear reply.
• Live events. Created, deleted, state, refused and policy changes stream as they happen.
• Metering built in. CPU, memory, disk, network, run and sleep time for every VM, ready for billing.

Runs on

• Linux x86_64
• Linux arm64
• macOS
• Windows x64
• Windows arm64

How it works

One daemon per host, one API, the same surface from the CLI and MCP.

CLI

flash host
flash host vms
flash host admissions
flash host events --follow
flash host usage --from 2026-11-01
flash host usage --owner alice --json

MCP

flash_host(method: "GET", path: "/v1/host")
flash_host(method: "GET", path: "/v1/host/usage?owner=alice")

API

GET  /v1/host
POST /v1/host/flashvms {"shape": {"vcpus": 2, "mem_mib": 1024}}
GET  /v1/host/events?since=0&wait_s=30
GET  /v1/host/usage?from=1793491200

The result

Every VM on a host is counted, admitted by fit and metered by the hour.

An idle VM holds as little as 27 MB of host memory, and flashhostd sees it all. Hosts run dense, and every VM is accounted for.