Skip to main content

Documentation Index

Fetch the complete documentation index at: https://langwatch.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

Returns the effective model list for the virtual key presented on the request. Entries are deduplicated and stable-sorted across three groups — OpenAI-compatible tools (Cursor, Codex, openai models list, etc.) treat the whole list uniformly.

​
Request

GET /v1/models
Authorization: Bearer lw_vk_live_<ULID>

​
Response

OpenAI-shape list object: 
{
  "object": "list",
  "data": [
    { "id": "claude",             "object": "model", "created": 1712000000, "owned_by": "anthropic"    },
    { "id": "gpt-4o",             "object": "model", "created": 1712000000, "owned_by": "azure_openai" },
    { "id": "gpt-5-mini",         "object": "model", "created": 1712000000, "owned_by": "openai"       },
    { "id": "claude-haiku-*",     "object": "model", "created": 1712000000, "owned_by": "anthropic"    },
    { "id": "anthropic/*",        "object": "model", "created": 1712000000, "owned_by": "anthropic"    },
    { "id": "openai/*",           "object": "model", "created": 1712000000, "owned_by": "openai"       }
  ]
}

​
The three groups

  1. Aliases from the VK’s model_aliases map — e.g. claude resolves to anthropic/claude-haiku-4-5-20251001, gpt-4o routes to the VK’s Azure deployment. owned_by reflects the resolved provider so UIs can render “gpt-4o (Azure)” correctly.
  2. models_allowed entries shown verbatim — including globs like claude-haiku-*. Clients that support pattern matching (most) pass these through; simpler pickers render them for the human to edit.
  3. Provider-type shortcuts — one entry per provider bound on the VK (openai/*, anthropic/*, bedrock/*, azure/*, vertex/*, gemini/*). Tools like Cursor that let the user type the exact model id can pick a concrete model under the shortcut (e.g. openai/gpt-5-mini) without waiting for the alias table to be updated. See Model Aliases → Aliases always win over explicit form for how the provider/model form interacts with aliases.
The list is deliberately permissive — it covers everything the VK could plausibly route, not a strict enumeration. models_allowed globs and provider shortcuts may not expand to real upstream models until you send a real request. The gateway’s per-request model resolution remains the source of truth for “will this actually route?”

​
Permissions

Only requires a valid VK; no extra permission. The returned list is scoped to what the VK can actually dispatch — engineers don’t need virtualKeys:view to call this.

​
Use cases

  • Model pickers in UIs that speak the OpenAI API — they call /v1/models and render the list.
  • Pre-flight check in CI: assert the CI’s VK has access to the models your tests depend on.
  • Debugging aliases: see what each VK exposes by hitting this with its token.
  • CLI probes: coding tools like Codex and Cursor call /v1/models on startup to auto-configure completion and model picking. The provider shortcuts (openai/*, anthropic/*) mean the user can type openai/gpt-5-mini directly in the picker without first adding an alias server-side. See Coding CLI integrations.

​
Errors

  • 401 invalid_api_key — unknown or malformed VK.
  • 403 virtual_key_revoked — VK revoked.
See API: Errors for the full table.