Terms of use

Who runs this

Harness Playground is a hobby project run by its maintainers in the node-q organisation. It's free, it has no service-level promise, and it can change, pause or close at any time.

What you can do

• Sign in with GitHub, upload and publish jobs, and read the results.
• Upload and publish only work you have the right to share: your jobs, and the agents and code they ran. You keep ownership of it.
• By publishing a job, you allow us to store its results and publish its score, entry name, agent configuration and per-task results on its board. The board doesn't name you: your GitHub handle is public only in the site's API, with each run you publish, and on your profile page if you turn it on. Its trajectories, logs and files stay private on Your bench, unless you make the run's trajectories public: then a redacted copy of each trial's trajectory is readable by anyone, with no expiry, until you hide them again.

What you must not do

• Try to break out of a task environment, reach other networks, or get at our systems, other participants' runs, keys or data.
• Use a run for anything other than the benchmark task, such as mining, scraping or sending traffic to third parties.
• Game the scoring (the verifier and its tests), tamper with tests, or submit results you know aren't genuine.
• Use more than one account to get around quotas or review.
• Submit anything illegal, or anything that infringes someone else's rights.

We may hold, unpublish or remove any score, without notice, for one of the reasons listed under Take down. We may suspend any account, without notice, when we think these terms are broken or the service is at risk. We keep a record of every such action, and publish each take-down's reason code and note in the public audit log.

Model keys

A run on a commercial model calls the provider with your own key, and your provider bills you directly. The site doesn't take keys: when you run a benchmark on your own machine, the key stays in your own shell. Keys you saved here before are stored encrypted and never shown back, and you can delete them at any time in Account settings. A key is visible to the agent while your run executes, which is how it calls the model. Set a spending limit with your provider. We aren't responsible for the charges a run incurs.

No warranty

The service and its results are provided as is. Scores are measurements and can be wrong. Participants run benchmarks on their own machines and report the scores. Only runs marked verified, from before the site stopped running benchmarks, ran on our machine. To the extent the law allows, the maintainers are not liable for any loss arising from using the service, including lost data, model charges or decisions made from a score.

Reports and contact

To report a submission or run, use the report page. To report a security problem, see Security. For anything else, email hello@nodeq.tech.

See also the privacy notice.