Models

The harness accepts only public model ids. Unknown ids return 404 model_not_found. Provider or upstream model names are intentionally absent from the list.

Public catalog

Discovery

Both compatible doors expose a GET /models route that returns the same catalog, so a host that follows either wire-compatible SDK pattern can discover models without parsing this page.

bash

curl https://harness.example.com/openai/v1/models \
  -H "Authorization: Bearer <VEDIKA_AGENT_KEY>"

Picking a model

Default: vedika-standard. Alias: vedika-intelligence.
Lower latency: vedika-swift (alias vedika-fast).
Cost-optimised: vedika-eco. Tools and web search are disabled at this tier.
Previous generation: vedika-standard-2.5 is retained as a stable alias.

Unknown ids fail before billing

A request with an unknown model id returns 404 and never reaches the billing stage. The same rule applies to tools or features that the selected tier cannot run: the server refuses them before settlement.

Aliases

Aliases are accepted wherever the canonical id is. They are listed in the table above. Retired names map to the successor model on a best-effort basis.