| # Terms of Use | |
| LLMAO, at https://llm.apache.org, is a service of The Apache Software | |
| Foundation's Responsible AI Initiative. By using it you agree to these | |
| terms. They are specific to this service and sit on top of your | |
| Individual Contributor License Agreement with the Foundation and the | |
| [[]Privacy Policy](https://privacy.apache.org/policies/privacy-policy-public.html), | |
| which continue to apply. | |
| ## Who may use it, and for what | |
| LLMAO is for official members of Apache Project Management Committees | |
| and ASF committers, and its use is limited to work on ASF codebases. | |
| ## What is recorded | |
| Every request sent through the gateway is logged. The log holds the | |
| prompt as the gateway received it (the messages, or the input of an | |
| embeddings request), the parameters sent with it, the model asked for | |
| and the one that served it, the key that was used and its owner and | |
| project, the tool that sent it, the token counts, the timing, and whether | |
| the request succeeded. | |
| Responses are not recorded. What a model produces is not kept. | |
| Logging is a condition of use. No request parameter or header switches | |
| it off; a request that tries is logged like any other. | |
| ## Who can read it, and why | |
| The service's site administrators, on the gateway host. No page on this | |
| site shows a prompt. The log exists so that we can understand how the | |
| service is used (which projects, tools and models, what shape the | |
| requests have, how that changes over time, how much capacity to run), | |
| and usage is monitored for non-ASF work, malicious use of AI, and | |
| criminal activity. | |
| ## How long it is kept | |
| Approximately 90 days, then it is deleted. The gateway host is covered | |
| by the Foundation's standard backups, which keep copies under their own | |
| retention. | |
| ## What to send where | |
| Treat anything you send as visible to the service's administrators. | |
| Models marked **Private** on the [[]Models page](/models) run on | |
| infrastructure the Foundation controls, and a prompt to one of them goes | |
| nowhere else. That is where unpublished security work belongs. Models | |
| marked **External** or **Mixed** send the request to a third-party | |
| provider under that provider's own terms; do not send them embargoed | |
| material. | |
| Whatever the model, do not send credentials, secrets, or personal data | |
| about other people. | |
| ## Keys | |
| A key is yours; do not share it. Use is bounded by the allowances shown | |
| on [[]My Keys](/keys), and a request that exceeds them is refused rather | |
| than queued. | |
| ## Changes and contact | |
| Development is ongoing. Updates are shared on general@rai.apache.org and | |
| in #llmao-notify on Slack, and the | |
| [[]Responsible AI Initiative](https://rai.apache.org/) site has more. | |
| These terms change when the service does. This page is the version | |
| running now, from | |
| [[]commit [commit]]([repo]/blob/[commit]/docs/help/terms.md); its history | |
| is in the [[]repository]([repo]/commits/main/docs/help/terms.md). |