You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Standard: run each unit's agent in a Boat VM of its own #149
extreme. A new backend for runs of the standard: where an agent's commands execute.
Components
app/nullius (backend)
app/web (routes)
templates and static files
docs and the site
the standard (ports, kits, relay, runs)
Today
A standard run's agent runs either as the user on this machine (local processes, with an explicit yes to running unsandboxed) or in a container, which needs Docker or Podman here. A machine without either has no sandboxed way to run an agent.
Wanted
A third backend, Boat: each unit's agent or capability runs in a boat.dev VM of its own, started from this machine with a key saved in Settings → Sandbox. The VM's only way out is the unit's relay on this machine, so the agent never holds a model key. The VM is deleted whatever ends the unit. The builder's Where step offers it with a VM size, and greys out what the account cannot start.
Acceptance
nullius standard run ... --backend boat [--boat-size SIZE] runs each unit's agent or capability in its own Boat VM, and conform takes --backend boat
The agent in the VM reaches only its relay; no model key is sent to the VM
Every VM a run starts is deleted, on success, failure, timeout and Stop
The VM's auto-stop follows the unit's time limit, within the 2 hours a free trial allows
Time spent starting the VM and installing the agent is not counted as the agent's time
Nothing under the user's own private folders (home, keys, Nullius config) is sent to a VM through a link
Settings → Sandbox saves, edits and removes the Boat key; Remove asks first
The builder's Where step offers Boat only on this machine, with the account's VM sizes; Review refuses a size the account cannot start
A run says when the account can start fewer VMs this hour or today than the run has units
Units that need a benchmark's MCP server or their own image end as not run on Boat, with the reason
docs/standard.md, docs/builder.md and docs/cli/run.md describe the backend
On Boat, a check that executes the agent's code (score.runs_code: HumanEval, the workspace kit) runs in this machine's throwaway check container, so such a run needs Docker here and is refused without it.
A unit whose setup gives the agent MCP servers (the tools kit, τ³-bench) is not run on Boat: they start from this machine's paths.
The Boat key is withheld from every other command Nullius runs, and a VM Boat will not delete is reported (delete_error in agent/boat.json).
Activity
aitianshi commentedon Nov 17, 2026
Scope update from the review of #150:
score.runs_code: HumanEval, the workspace kit) runs in this machine's throwaway check container, so such a run needs Docker here and is refused without it.delete_errorinagent/boat.json).feat(standard): run each unit's agent in a Boat VM of its own (ThinkF…