Fast cluster queries for people. Structured SAI operations for scripts and Agents.
Use familiar SSH commands on a cluster, or connect your own computer and Agent through normal SAI authorization. The client contains no model loop and does not require Pilot.
sqm # your jobs
sqa # the applied computing account and its child projects
gpu # GPU allocation and available nodes
cpu # CPU and memory allocation
quota # quota for the current directory
sai tui # select jobs/nodes with arrows or a click; q exitsThese queries run locally under your Unix identity. In the TUI, p pauses or
resumes updates, Enter inspects a row, and c requests that job's running cost
through your authorized SAI account. Allocation is distinct from utilization;
running estimates are distinct from settled charges.
The job selector discovers your native Slurm accounts; quota lets you choose an
actual directory. Job details lead with the allocation and a compact CPU/NUMA
and GPU connection view when a current hardware record is available. i expands
hardware details; arrow/page keys scroll, r refreshes, and Escape returns to the
selected row. Native account names are also available as
JSON through sai account local. These reads do not change your shell or defaults.
Native adapters have been exercised with Slurm 26.05.1 and Lustre 2.17; other native versions are not yet qualified.
Python 3.11+ is required. This alpha is qualified on Linux; macOS and Windows are not yet qualified. Download the wheel and SHA-256 checksum from the versioned releases, verify the checksum, then install in an isolated environment with pipx:
pipx install ./sai_cli-0.2.0a2-py3-none-any.whl
sai configure https://console.example.org --format table
sai auth login
sai sites --format tableUse your service's trusted URL in place of the example. Confirm the login code in your browser, then choose a returned site explicitly:
sai resources catalog --site site-a --format table
sai account defaults --site site-a --format tableThe catalogue contains published specifications and prices. Account defaults shows your authorized choices; neither command changes an account or grants site access. Configuration preserves existing profiles and credentials. See first use and recovery for wheel installation, additional permissions, multiple services and upgrade/rollback.
The base client has no third-party runtime dependency. To add the optional
interface, including to an existing base installation, use
pipx install --force './sai_cli-0.2.0a2-py3-none-any.whl[tui]'. This preserves
your user configuration and credentials. It needs an interactive
terminal. Local views require the corresponding Slurm/Lustre tools on that host.
The 0.2.0a2 candidate includes recorded hardware connections, CPU locality and
selected-site connection discovery. Use the wheel filename from the reviewed
release you downloaded; older release assets are retained unchanged.
sai describe
sai describe account set-default
sai skill --format markdowndescribe is the installed command authority. skill exports matching Agent
instructions for the host you choose; neither requires login or network access.
Commands return JSON by default, with separate structured errors and documented
exit codes. A task can request its missing permissions with
auth login --for, followed by the user's browser confirmation.
The client supports native queue/resource/quota reads, account and cost queries,
job allocation and recorded-run outputs, default-account changes and recipient
invitation acceptance. Broader project, storage and compute management is not
implemented. Read permission does not grant write permission. The called service
owns authorization and reports whether a change was accepted or applied.
invitations accept ID previews file and computation permissions and the payer;
add --apply to accept. This v2 workflow requires the paired Portal update.
- First useful task
- Human and Agent scope
- Supported management and receipts
- Architecture and ownership
- Versioned API contracts
- Contributing and security reporting
- Licensing and third-party notices
First-party client code, the shared native projection and bundled Agent guidance are licensed under Apache-2.0. The separate Slurm source patch retains its upstream GPL terms and is excluded from Python distributions. The repository's site integration artifacts have their own deployment roles; installing the Python client never installs or changes a scheduler, login profile or storage service.
The interconnect contract joins recorded CPU/NUMA, PCIe, GPU and fabric
relationships to an exact live allocation via --with-interconnect. Local reads
stay native; the existing site gateway uses include_interconnect=true.
Deployment and its validation boundary are recorded in
terminal scope.