Terms of Service

By creating an account or using the service you accept these terms. If you do not, do not use it. The service is operated by Abuelgasim Mohamed Ibrahim Adam as an independent research project.

Tools for measuring the operational reliability of AI agents: evaluation against a fault battery, live monitoring, policy-driven halting, alerting, and signed behavioural records.

Plans: the free tier allows 500 evaluations a month and 3 agents. Pro allows 5,000 evaluations a month. Quotas are shared across your agents and reset monthly. Monitoring sessions do not consume evaluation quota.

This section matters more than most, so it is stated directly rather than buried in a disclaimer.

• No uptime guarantee. A single instance in one region, no redundancy, no automatic failover. Service is best-effort.
• No support commitment. No response-time obligation, no on-call, no account management.
• No compensation for downtime, and no service credits.
• No certification. HB-Eval is not an accreditation body. An Agent Passport records observed behaviour and is signed; it is not a certificate, an audit, or a warranty about anything your agent will do next.
• No guarantee of correctness. The metrics measure behavioural reliability under fault. They do not judge whether your agent’s answers are true.

The status page states the same commitments and publishes what the service is actually doing.

You retain all rights to your agents, your tasks, your prompts and your results. We claim no ownership and no licence beyond what is needed to operate the service for you.

We do not use your data to train models. If you opt in to the Observatory, anonymised aggregate figures — never content, never identifiers — contribute to public statistics.

Export and deletion are self-service and immediate, in Settings.

• Keep your API keys and agent secrets confidential. Rotate them from the agent page if exposed.
• Do not attempt to exceed your quota by creating multiple accounts, and do not attempt to disrupt the service for others.
• Do not use the service unlawfully, or to process data you have no right to process.
• You are responsible for what your agent does. Safe Halt raises a flag your code must act on; it stops the next step and cannot undo a previous one.

A passport is a signed record of behaviour observed by this service, valid for 90 days from issue. Publishing one is your choice, produces a public link, and can be withdrawn at any time.

You may share and republish passports we issue to you. You may not alter one and present it as issued by us — the signature makes any alteration detectable, and doing so would be misrepresentation.

Passports marked demo: true record a scripted demonstration on this site and describe no deployed agent.

The service is provided “as is” and “as available”, without warranties of any kind, express or implied.

To the fullest extent permitted by law, the operator is not liable for indirect, incidental or consequential damages, for lost profits, or for loss of data. Total liability for any claim is limited to what you paid in the twelve months before it arose — which, on the free plan, is nothing.

Nothing here excludes liability that cannot lawfully be excluded.

You may delete your account at any time, which removes your data as described in the privacy policy.

We may suspend an account that is disrupting the service or being used unlawfully. Where circumstances allow, notice will be given first and an export offered.

As a research project, the service may be discontinued. If that happens, reasonable notice will be given so that data can be exported.

These terms may change; the date above will reflect it. Continuing to use the service after a change means you accept it.

abuelgasim.hbeval@outlook.com