CLI (hangar)

pip install ./cli            # from a clone; requires Python 3.10+
hangar login http://localhost:8090 --token <personal token>   # UI → Chaves de API → scope "user"

Credentials are read from HANGAR_URL / HANGAR_TOKEN, or from ~/.config/hangar/config.json (written by login, mode 600). Every command accepts --json.

Command	What it does
`hangar whoami`	Your company role and teams
`hangar agents ls [--mine]`	Agents you can see: team, your access, status, version, stage/prod, requests and cost (7d)
`hangar agents get <slug> [--spec]`	Detail, or just the spec as YAML
`hangar apply -f file.yaml [--ship]`	GitOps apply (`{skills, mcp_servers, agents}`; multiple YAML documents allowed; `-` = stdin). Template files are accepted too
`hangar export <slug>`	Print an agent as an applyable YAML document
`hangar test <slug>`	Run stage tests; exit code 1 if they fail (useful in CI)
`hangar ship <slug>`	Stage → tests → prod
`hangar deploy –env stage	prod`
`hangar rollback <slug> <version>`	Restore an old spec as a new version
`hangar chat <slug> <message…> [--env]`	Talk to an agent
`hangar logs <slug> [--env]`	Container logs
`hangar jobs run <slug> <task…> [--follow] [--timeout N]`	Harness job; `--follow` streams status and logs, Ctrl+C cancels
`hangar jobs get <id>` / `hangar jobs cancel <id>`	Inspect / cancel
`hangar templates ls` / `hangar templates apply <id> [--connection C] [--harness-connection H]`	Template gallery
`hangar connect <slug> [tool] [--mode mcp\|model] [--preview] [--list]`	Plug an agent into a tool: creates a per-tool key and prints the config (steps on stderr). No tool = list tools
`hangar keys ls` / `create <name> --scope invoke --agent <slug>` / `revoke <id>`	API keys (`create` prints the key on stdout). `--scope user` = personal token for the CLI and the platform MCP
`hangar teams ls` / `hangar teams add <team> <email> --role developer`	Teams (with spend) and members
`hangar approvals ls` / `approve <kind> <id>` / `reject <kind> <id>`	Production (`promotion`) and access (`access`) approvals
`hangar request-access <slug> <reason…>`	Ask a team to use its agent
`hangar schedules ls [slug]` / `add <slug> <message…> --cron '0 9 * * 1-5'` (or `--at 2026-10-05T09:00`) / `run <id>` / `rm <id>`	Schedules: the agent runs by itself in production

Install and diagnose

./scripts/install.sh                     # installs the CLI in .hangar/venv and runs `hangar setup`
hangar setup                             # guided install / reconfigure (Docker)
hangar setup --answers setup.yaml        # unattended (see setup.example.yaml); --no-start only writes .env
hangar doctor                            # checks the installation; --json for monitoring
hangar cleanup                           # removes data volumes of agents already deleted (asks first)
hangar setup --target kubernetes         # install / upgrade on a cluster with Helm (docs/kubernetes.md)
hangar doctor --namespace agent-hangar   # pods + every check above, via a temporary port-forward
hangar setup --configure-only            # cluster installed by Terraform/GitOps: only LLM, admin and AI tools (no helm)
hangar backup                            # pg_dump to .hangar/backups/ (--namespace for the chart's Postgres)
hangar restore <file.dump>               # restores (saves the current state first)
hangar upgrade --version 0.12.0          # Docker: backup, new images, start, doctor (docs/backup.md)

hangar test-drive runs the whole path against your installation, with the test model (no LLM, no cost): a team, a Digital employee, its probation, the admission, one task and the result, with the portal links. It deletes what it created at the end; --keep leaves it there to look at, --manager <e-mail> picks the manager (default: you, or a test person with the admin token).

hangar test-drive            # ✓ team · ✓ hired · ✓ probation · ✓ admitted · ✓ task done → links
hangar test-drive --keep     # keep the team and the Digital employee to open them in the portal
hangar test-drive --llm openrouter   # the same path with a real model (an LLM connection from the catalog)

hangar demo creates an example company so a new install doesn’t open empty: the team Exemplo · Renovações with you as manager, a goal, a project and three Digital employees in an org chart (the coordinator and two who report to her), already admitted, and a first task. Without an LLM they use the test model; --llm <connection> uses a real one. Running it again doesn’t duplicate anything. hangar setup offers it at the end (demo: true in the answers file).

hangar demo                  # ✓ team · ✓ hired · ✓ org chart · ✓ admitted · ✓ project and first task → links
hangar demo --remove         # deletes the example (the project is archived)

See install.md.

Agents in CI

- run: pip install ./cli
- run: hangar apply -f agents/ --ship
  env:
    HANGAR_URL: ${{ secrets.HANGAR_URL }}
    HANGAR_TOKEN: ${{ secrets.HANGAR_ADMIN_KEY }}

apply only creates versions for specs that changed, and ship refuses to promote a version whose tests fail.