# Noach Ark

> Find GPU rentals for a model and concurrent active users, with sourced memory estimates and whole-instance compute costs.

## Agent access
- [Instructions](https://noachark.com/for-agents/): Read the workflow, assumptions and limits before making recommendations.
- [OpenAPI](https://noachark.com/api/v1/openapi.json): Read-only GET API; no authentication required.
- [Models](https://noachark.com/api/v1/models/): Discover exact model slugs and sizing support.
- [GPU catalog](https://noachark.com/api/v1/gpus/): Optional technical requirements path.
- [Rental search](https://noachark.com/api/v1/rentals/search/): Model is the default mode. Supply a model slug; parallel_users defaults to 1.
- [Human finder](https://noachark.com/find/): The same eligibility and cost calculations.

Use the JSON API instead of inferring prices from snippets. Echo defaults and cite evidence.
Estimated memory fit does not establish throughput or latency. Costs are partial estimates, not complete quotes.
An empty result means no eligible observed offers; do not silently change the user's requirements.