Things you run
| word | meaning |
|---|---|
| Job | a Python function that runs in a container: .remote() once, .map() fan-out, .remote() from another function to build pipelines |
| Endpoint | a function behind an HTTP URL, invoked synchronously; @asgi wraps a whole ASGI app, @realtime a WebSocket handler |
| Queue | background work, one task per item, with retries; .put() from Python or POST from anywhere |
| Cron | a function on a cron expression (UTC) |
| Box | any long-running process with ports, for servers, databases and notebooks |
| Sandbox | a box you drive from Python: run code and processes, read and write files, expose ports, snapshot memory |
| Task | one execution of any of the above; it has a status (pending, running, complete, error), a container and logs |
| Container | the process running a task. A container that is already up answers immediately; one that has to start pays for the blueprint, the container setup and whatever the process loads - see Cold starts |
Things you attach
| word | meaning |
|---|---|
| Blueprint | what is installed in the container: Python version, pip and apt packages, shell commands. Built once and content-addressed, so a changed line rebuilds only from that step |
| Volume | persistent files shared between containers, browsed in the dashboard or with levers9 ls / cp / rm; a CloudBucket mounts S3 or GCS the same way |
| Secret | a value stored in the workspace and injected as an environment variable into containers that ask for it |
| GPU | gpu="L40S" or a preference list; the scheduler takes the first type with free capacity. gpu_count=2/4/8 gives one job several cards on one node for one training with levers9.distributed; .map() over gpu_count=1 jobs runs many trainings side by side. See Train on 1, 2, 4 or 8 cards |
| Pool | a group of workers with the same hardware (default, build, l40s); a box or job can be pinned to one |
| Cluster | a Kubernetes cluster you run, attached as a pool with levers9 cluster join: one agent inside it dials out, reports the GPU nodes, and starts workers when asked; nothing dials in. See Attach a cluster |
Things you ship
| word | meaning |
|---|---|
| Deployment | a named, versioned thing you can call: /job/NAME/v1, /endpoint/NAME/v2, /queue/NAME/v1, /box/NAME/PORT/... (the older /function, /taskqueue and /pod paths still answer) |
| Version | every levers9 deploy makes one; older versions keep working until stopped |
| App | the group of functions and deployments in one script, shown together in the dashboard; deployed, stopped or ephemeral (ran once, never deployed) |
| Serve | a temporary URL for the code in front of you, with hot reload, gone when you stop it |
Who you are
| word | meaning |
|---|---|
| Workspace | a team: its apps, volumes, secrets and members. Everyone with the same email domain shares one; you can create more for a project or a client |
| Environment | a separate space inside a workspace (main, staging, a branch). Each is a controlplane workspace of its own, so nothing leaks between them |
| Token | what the CLI and the dashboard use to act as one environment; levers9 login fetches it, levers9 token manages extra ones |
| Context | one entry in ~/.levers9/config: controlplane host, port and token; switch with -c NAME |
cpu, memory, gpu, image, volumes, secrets, env,
timeout, retries. They are listed on Parameters.