Piotr Mlocek
96e9945cfd
## What changed - Added tag-driven release flow in GitLab CI: - new `release` stage with tag-only jobs - `publish_tag_artifacts` publishes Docker + Python artifacts on `vX.Y.Z` tags - `create_release_notes` generates release notes from conventional commits via `git-cliff` and creates a GitLab release via `glab` - Updated main-branch image publishing to version-aware tagging: - `publish_ecr_images` now runs `mise run publish:main` - main publishes `:dev`, `:latest`, and a versioned dev tag - Added release-oriented mise tasks: - `publish:main` - `publish:tag` - `python:publish:macos` (manual macOS arm64 wheel publish) - Switched Linux Python wheel builds to buildx: - added `deploy/docker/Dockerfile.python-wheels` - replaced old per-arch docker-run tasks with `python:build:multiarch` - Added macOS arm64 wheel build path for local publishing: - `python:build:macos` builds `aarch64-apple-darwin` - intended to run locally on macOS after tag CI finishes - Made Docker multiarch publish script tag-flexible: - `TAG_LATEST` is no longer hardcoded in ECR mode - supports `EXTRA_DOCKER_TAGS` - applies extra tags to sandbox/server/pki-job/cluster images - Moved release tooling to `build/scripts/release.py` and updated all mise references - Removed obsolete Docker Artifactory env vars from `mise.toml` ## Release behavior - **Main branch CI** - Docker: `:dev`, `:latest`, and versioned dev tag - **Tag CI (`vX.Y.Z`)** - Docker: `:X.Y.Z` only (no `:latest`) - Python: Linux wheels published from CI - GitLab release notes created from conventional commits - **Manual macOS step (after tagging)** 1. Checkout the tag locally on macOS 2. Run `mise run python:publish:macos` ## Validation - `mise run python:lint` - `mise run version:print` - `mise run python:build:macos` - `uv run python build/scripts/release.py --help`
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
🚧 Under construction 🚧
uv pip install navigator --index-url https://urm.nvidia.com/artifactory/api/pypi/nv-shared-pypi/simple
navigator sandbox create -- claude # or opencode or codex
The sandbox container includes a minimal networking toolset by default: ping, dig, nslookup, nc, traceroute, and netstat.
Developing
See CONTRIBUTING.md for more information about developing and contributing to Navigator.
Architecture
See architecture/ for detailed architecture docs and design decisions.