Community model API

Get access

Sign in with LinkedIn on Apply. Once approved, open API management from the workspace to generate a token. Use the same token for model requests, sent as Authorization: Bearer. New tokens last 30 days. Existing tokens keep their original expiry. Regenerating your token revokes the previous one. Keep the token out of source code, logs, and URLs.

If you use Python urllib, send a descriptive User-Agent such as my-app/1.0. Cloudflare rejects the library's default Python-urllib/... signature before the API can check your token. Python requests already sends an accepted User-Agent by default.

List models

GET /v1/models returns every available model with its exact name, source links, runtime, license, and GPU. Use your application token to call it.

curl https://genaicommunity.ai/v1/models \
  -H "Authorization: Bearer $COMMUNITY_API_TOKEN"

Make a decision

Send the state you want evaluated and a map of typed questions. The required model field selects which model to call.

curl https://genaicommunity.ai/v1/systemone \
  -H "Authorization: Bearer $COMMUNITY_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mys/laya-typed-decisions-GGUF",
    "state": {"message": "I was charged twice for the same invoice."},
    "questions": {
      "billing": {
        "type": "noul",
        "instructions": "Is this a billing issue?"
      }
    }
  }'

For multilingual input, use the same request with "model": "mys/laya-multilingual-GGUF" and put your text and question instructions in the desired language.

For Hopper, set "model": "HopitAI/hopper" and send one question. Its answer uses the same top-level answers structure.

For Xor, set "model": "juspay/xor", supply state as text, and use an ordered array of 2 to 26 level descriptions for a score question. Optional images accepts up to eight static PNG, JPEG, or WebP base64 data URLs. Images are limited to 5 MiB decoded and 16 megapixels combined; the full JSON request is limited to 8 MiB.

The response includes top-level model, answers, and usage, matching the SystemOne response convention. Usage includes input_tokens and output_tokens plus our GPU extensions: gpuSeconds for this call, totalGpuSeconds for your account, and requestCount.

Usage includes the GPU type and count, its seconds unit, and measurement: "inference". GPU seconds equal measured inference time times GPU count. A completed model call is metered even if the model rejects its input. A cold Xor app may return 503 model_starting; retry after the Retry-After interval.

Read your usage

curl "https://genaicommunity.ai/v1/models/usage?model=mys%2Flaya-typed-decisions-GGUF" \
  -H "Authorization: Bearer $COMMUNITY_API_TOKEN"

Replace the model query value with mys%2Flaya-multilingual-GGUF, HopitAI%2Fhopper, or juspay%2Fxor for separate totals. GPU seconds measure elapsed inference time inside the allocated GPU container. Startup, time waiting outside the container, and idle time are excluded. This meter does not measure GPU utilization or billed usage. Your totals belong to your member account, so they stay the same when you renew your token.