ComputeRadar catalog API
The normalized catalog is queryable without loading the browser interface. Requests read the local synchronized index, so provider API latency is not part of catalog response time.
Endpoints
GET /api/catalogServer-side search, filters, sorting and paginationGET /api/providersProvider connector and last-sync statusGET /api/statsCatalog summary statisticsGET /api/healthLocal catalog / sync healthGET /feed/offers.jsonOffers-only JSON feedGET /feed/offers.csvCSV exportGET /openapi.jsonOpenAPI 3.1 descriptionFilters
q, type, provider, gpu, cpu, country, city, location, storageType, currency, variant, pricingBasis, available, hasGpu, gpuOptional, hasPrice, min/max hourly and monthly price, RAM, CPU/vCPU, CPU threads, VRAM per GPU, total GPU VRAM, GPU count, storage, download/link bandwidth, upload bandwidth, traffic, reliability and stock quantity, plus maxAgeMinutes.
Sorting
Use price-hour-asc, price-hour-desc, price-month-asc, price-month-desc, ram-asc/desc, cores-asc/desc, threads-asc/desc, vram-asc/desc, total-vram-asc/desc, gpu-count-asc/desc, storage-asc/desc, bandwidth-asc/desc, upload-asc/desc, traffic-asc/desc, reliability-asc/desc, quantity-asc/desc, updated-asc/desc, provider-asc/desc, gpu-model-asc/desc, cpu-model-asc/desc, location-asc/desc or title-asc/desc.
Pagination
page starts at 1. limit accepts 10–200 rows per request. Responses include matchingOffers and totalPages.
Examples
/api/catalog?type=gpu&gpu=H100&available=1&sort=price-hour-asc&page=1&limit=50
/api/catalog?type=gpu&gpu=RTX%204090&maxHourly=0.50&sort=price-hour-desc
/api/catalog?type=bare-metal&minRam=512&minCores=64&available=1&sort=price-month-asc
/api/catalog?cpu=EPYC%209754&country=United%20States&minBandwidth=1000
/api/catalog?minGpus=4&minVram=80&minReliability=0.98&sort=reliability-desc
Freshness
Provider APIs update independently in the background. Responses include the catalog timestamp, source status and per-offer updatedAt.