Drew Newberry
aa886bedb9
After fork_start(), the parent now: - Polls https://localhost:<kube_port>/readyz with 2s intervals (120s timeout) - Checks child PID is alive between polls (fast-fail on VM crash) - Reads kubeconfig from rootfs via virtio-fs (host-visible) - Rewrites server URL and cluster name, stores in standard location - Prints kubectl usage instructions when ready Always forwards kube API port (ephemeral if --kube-port not specified) to enable health checking. Adds --name flag for kubeconfig context naming. Boot reliability: vm-init.sh now retries k3s up to 3 times with cleanup between attempts, handling the transient kine SQLite race condition on tmpfs that occasionally crashes k3s on first boot. Exports rewrite_kubeconfig and store_kubeconfig from navigator-bootstrap for use by the gateway cluster command.
Navigator 🗺️
Navigator is the runtime environment for autonomous agents—the "Matrix" where they live, work, and verify.
While coding tools like Claude help agents write logic, Navigator provides the infrastructure to run it, offering a programmable factory where agents can spin up physics simulations to master tasks, generate synthetic data to fix edge cases, and safely iterate through thousands of failures in isolated sandboxes.
It transforms the data center from a static deployment target into a continuous verification engine, allowing agents to autonomously build and operate complex systems—from physical robotics to self-healing infrastructure—without needing a human to manage the infrastructure.
Quickstart
Prerequisites
- Docker — Docker Desktop (or a Docker daemon) must be running.
- Python 3.12+
- uv 0.9+
Install
uv pip install nemoclaw \
--upgrade \
--pre \
--index-url https://urm.nvidia.com/artifactory/api/pypi/nv-shared-pypi/simple
The navigator binary is installed into your Python environment. Use uv run navigator to invoke it, or activate your venv first with source .venv/bin/activate.
Create a sandbox
To install a Openclaw cluster and start a sandbox
navigator sandbox create -- claude # or opencode or codex
To run a sandbox on a remote machine, pass --remote [remote-ssh-host].
For more information see navigator sandbox create --help.
The sandbox container includes the following tools by default:
| Category | Tools |
|---|---|
| Agent | claude, opencode, codex |
| Language | python (3.12), node (22) |
| Developer | gh, git, vim, nano |
| Networking | ping, dig, nslookup, nc, traceroute, netstat |
For additional sandbox images see the NVIDIA/NemoClaw-Community images. You can also build your own custom images.
Deploy a cluster
Note: navigator sandbox create automatically deploys a cluster if one isn't already running.
To deploy a cluster explicitly:
navigator cluster admin deploy
For remote deployment:
navigator cluster admin deploy --remote user@host
Upgrading
To upgrade, redeploy your cluster to pick up the latest server and sandbox images:
navigator cluster admin deploy
This will prompt you to recreate the cluster. Select "yes" to recreate the cluster.
Developing
See CONTRIBUTING.md for building from source and contributing to Navigator.
Architecture
See architecture/ for detailed architecture docs and design decisions.