Docs

Run your first Job

A Job is one run you start; each attempt inside it is a Trial. This walkthrough runs a complete Job from the command line and shows you how to read and repeat it.

It uses the support queue project, a small help desk where an AI sorts tickets into categories. Its scripted Agent follows keyword rules instead of calling a model, so every command here until Record it in the web app works offline, costs nothing, and needs no account.

You need Plural installed; see Install.

Validate and run

Unpack the project, check it, preview the run, then run it:

bash

unzip first-project.zipcd first-projectplural benchmark validate support-triageplural run --benchmark support-triage --agent scripted --dry-runplural run --benchmark support-triage --agent scripted

validate checks the Benchmark and every Task, Verifier, Environment, and Harness it depends on.
--dry-run prints the plan and each input's version and content hash (a fingerprint of its exact files) without running anything.
The real run creates a Job with one Trial per Task, prints each Trial as it finishes, and ends with the Agent's mean score:

text

Running benchmark/support-triage with scripted (openai/gpt-5.6-luna) locally: 3 Trials, 3 at a time (auto: every Trial at once).  [1/3] trl_53c727d33aeaa097dda21e40  ticket-2  succeeded  score=1.000  | 1 succeeded, 0 failed, mean 1.000  [2/3] trl_5fd546be688383450e6dd8d2  ticket-3  succeeded  score=1.000  | 2 succeeded, 0 failed, mean 1.000  [3/3] trl_d124de45ec874ebc651f51dc  ticket-1  succeeded  score=1.000  | 3 succeeded, 0 failed, mean 1.000Job job_7ca838103054ee8c4ae74a51 succeeded.  scripted: mean score 1.000 over 3 Trials, 3 succeeded, coverage 100%Details: plural job show job_7ca838103054ee8c4ae74a51

Trials run at the same time, so they can finish in any order. The model shown is recorded on the Agent but never called.

Inspect and rerun

Every run is a new Job, recorded under .plural/jobs/<job-id>/ with its inputs pinned by version and content hash. Use the ids the run printed:

bash

plural job listplural job show <job-id>plural trial show <trial-id>

job show lists each Trial with its Task, status, and score.
trial show prints the score, each Verifier's evidence, and where the Trial's artifacts and logs are stored.

To run the exact same experiment again:

bash

plural job rerun <job-id>

A rerun uses the exact inputs the original Job pinned, not your current files, and creates a new Job linked to the original. That means you can edit the project freely and still repeat an old result.

Record it in the web app

Every Job runs from your computer. To see it in the web app and share it with your team, sign in, register the project, and add --record. This step needs a Plural account:

bash

plural auth loginplural project init support-queue --pushplural run --benchmark support-triage --agent scripted --record

plural auth login opens your browser to sign in and stores a 30-day API key, which covers both hosted commands like these and model calls from your own computer; see Sign in. --record first pushes whatever the run uses that the hosted project lacks, then records each Trial as it runs. To run the Trials in cloud sandboxes instead of on your computer, change the Environment's Runtime; see Track runs and run in the cloud.

The credential is stored in your user config folder, never in the project. The hosted project is private, and pushing never makes anything public. Follow the Job in the web app under Jobs.

The support queue tutorial explains each file in the project and how to read the results.