CLOUD AGENTS / LAUNCH

Start a cloud agent from the terminal

Launch agents in the cloud with plain-English descriptions. The agent runs in an isolated cloud sandbox, has access to your repo via read-only tools, and reports metrics and optionally traces back to the dashboard.

How it works

Cloud sandbox

Each agent runs in its own persistent Linux VM. The sandbox is forked from a snapshot in seconds, stopped between runs to keep compute near zero, and resumed on demand.

Our harness

A hosted nah session runs our harness, never preinstalled agents. The sandbox runs our code, our tools, and our event stream — so the UI and the runtime stay consistent.

When you launch an agent, Nah Cloud creates a persistent Linux cloud sandbox, installs the nah runtime, and starts the agent loop. The agent connects back to the dashboard over a secure channel, streaming events as they happen.

Launch from the terminal

Log in once, then launch agents with a plain-English description:

bash
# Authenticate once — stores a bearer token scoped to your tenant
nah cloud login

# Launch an agent from a plain-English description
nah cloud launch "Review the login flow in src/auth.ts and summarize failure cases"

# Check the status of running agents
nah cloud status

# Stop an agent
nah cloud stop <agent-id>

# View the dashboard URL for this agent
nah cloud url <agent-id>

Environment variable fallback

If you prefer not to use the login flow, setNAH_CLOUD_TOKEN as an environment variable — useful for CI or scripting.

Launch parameters

When launching programmatically, you can pass the full parameter set:

bash
curl -X POST https://api.nah.cloud/v1/agents/launch   -H "Authorization: Bearer $NAH_CLOUD_TOKEN"   -H "Content-Type: application/json"   -d '{
    "prompt": "Review the login flow in src/auth.ts and summarize failure cases",
    "model": "anthropic:claude-sonnet-4",
    "permissions": "ask",
    "trace": false,
    "sandbox": { "type": "small", "ttlSeconds": 3600 }
  }'
prompt
Plain-English task description
model
Provider:model
permissions
ask | yolo | readonly
trace
false = metrics only, true = full traces
sandbox.type
small | large
sandbox.ttlSeconds
Auto-stop after N seconds

After launch

Once launched, the agent connects to the Nah Cloud dashboard where you can:

  • Monitor progress in real-time via the event stream
  • View token usage and step counts as they accumulate
  • Stop the run at any time — billing stops immediately
  • Open the trace waterfall if tracing was enabled

The dashboard URL is printed to the terminal on launch, or retrieve it withnah cloud url <agent-id>.