4. Your content & data handling
Ownership. "Customer Content" means the prompts, inputs, files, datasets, container images, and code you submit, batch inputs and results, fine-tuned model artifacts created for you, and model outputs processed through the Services. As between the parties, you own your Customer Content. You grant Lilac the limited right to host and process Customer Content solely to provide, secure, and support the Services. Lilac does not use Customer Content to train models and does not sell Customer Content.
Data handling by Service type. Retention differs by Service:
- Real-time inference API: request and response bodies are processed transiently in memory; Lilac does not store prompts or completions at rest after the response is returned, as further described in the Privacy Policy.
- Batch, container, training & storage Services: by their nature these Services store the Customer Content you submit and the results they produce until you retrieve or delete them. Unless Service-Specific Terms state otherwise, stored job inputs and results are retained for 30 days after job completion and then deleted, and you may delete them earlier via the Services.
- Operational metadata (for example, token counts, job and workload identifiers, resource usage, latency, and error codes) is retained for billing, reliability, security, and abuse prevention.
Capacity Providers. Workloads may execute on Capacity Provider infrastructure. Capacity Providers are contractually prohibited from retaining, inspecting, or using Customer Content, and process it only transiently to execute your workloads.