AI models, limits and data policy

The catalogue lists the models Sylphx AI serves. The model pages show it in full — context window, maximum output, data policy and a ready-to-copy request for every model — and read it from the same catalogue this API serves, so the page and the API cannot disagree.

#Listing models

Two doors serve the catalogue: one for your key, one public.

curl https://api.sylphx.ai/v1/models \
  -H "Authorization: Bearer $SYLPHX_API_KEY"

# One model document:
curl https://api.sylphx.ai/v1/models/openai/gpt-5.5 \
  -H "Authorization: Bearer $SYLPHX_API_KEY"

curl https://api.sylphx.ai/v1/public/models

# One public model, including its description:
curl https://api.sylphx.ai/v1/public/models/openai/gpt-5.5

#Switching models

Because model is a field, a switch is a one-line change. The request and stream shape stay identical.

{
  "model": "qwen/qwen3.8-flash",
  "input": "Summarise this incident report in three bullets."
}

Read the new model's page first: context window, maximum output and data policy can differ. A conversation that used provider-sealed state should stay on the family that produced it; see switching models safely.

#Data policy

Each catalogue row publishes the default routing posture for that model: whether API data may be used for training, and whether zero data retention (ZDR) applies. ZDR is published only for lanes that carry an evidenced provider mark (a provider's ZDR endpoint listing or a reviewed lane declaration). A lane without that mark is shown and served as standard: it is not refused, and it is not claimed as ZDR.

• train and zdr are booleans describing that default. zdr is true only when every default lane has an evidenced ZDR mark, and false otherwise.
• posture is standard or consent_trained; source is model_declaration or default; default_routing is standard.
• tiers states which of standard, zdr and training the model has at least one lane for. tiers.zdr counts only evidenced ZDR lanes.
• training_opt_in is true when lower-cost options that may train on data exist for the model. They are reached only when the caller asks for them with Sylphx-Data-Policy: training; without that, a request never uses them.
• Every response from a routed lane carries Sylphx-Data-Policy: zdr | standard | training, the posture of the lane that actually served it. A response served by a lane that may train on data says training; a lane without an evidenced ZDR mark says standard, even when you asked for zdr.
• The posture belongs to the model row, not to your key, so switching models can change it. Check it before you move a sensitive workload.
• Retention of stored responses and files is separate; see the privacy policy.