Use Claude Desktop with InferAll: third-party inference setup

Claude Desktop can run on a third-party inference gateway instead of a claude.ai sign-in. InferAll implements the Anthropic Messages API, so the app can use it. This page gives the settings that work, based on Anthropic's configuration reference and our own checks against the live API on 2026-09-25.

Settings

Open Developer → Configure Third-Party Inference… in Claude Desktop (turn on Developer Mode first if you do not see a Developer menu). Then set:

Setting	Value
Inference provider	gateway
Gateway base URL	https://api.inferall.ai/v1
Gateway API key	your ifu_... key from inferall.ai/keys
Gateway auth scheme	bearer (the default) or x-api-key; both work

Choose your models

Model discovery works (updated 2026-09-26). With discovery on, the app fills its menu from our /v1/models, which now lists every chat model by the exact id the gateway routes, for example anthropic/claude-opus-4-8 and nvidia/nemotron-3-super-120b-a12b. Pick from that menu.

If you prefer a short menu, turn discovery off and add models by hand in the Models section, using these ids:

Model id to add	Price
nvidia/nemotron-3-super-120b-a12b	$0 (free tier)
nvidia/nemotron-3-ultra-550b-a55b	$0 (free tier), our largest free model
anthropic/claude-opus-4-8	paid, billed from your balance
anthropic/claude-haiku-4-5-20251001	paid, billed from your balance

Other Claude, GPT and Gemini models work the same way with an anthropic/, openai/ or gemini/ prefix. The live list, in exactly the ids to use here, is at GET https://api.inferall.ai/v1/models (send your key).

Configuration takes effect at launch, so fully quit and reopen Claude Desktop after any change.

What we checked (2026-09-25, discovery re-checked 2026-09-26)

• POST https://api.inferall.ai/v1/messages answers with either Authorization: Bearer or x-api-key.
• GET https://api.inferall.ai/v1/models answers with a bearer key.
• anthropic/claude-opus-4-8, anthropic/claude-haiku-4-5-20251001 and nvidia/nemotron-3-super-120b-a12b each answered as the model named.
• On 2026-09-26, /v1/models listed 8 Claude ids, all prefixed anthropic/; the ones we called answered on /v1/messages as the model named.

We have not tested every Claude Desktop feature against the gateway. If something in the app does not work, email contact@kindly.fyi with the model id and what you clicked.

Paid models and the free trial

New accounts get 25 free calls on open models, no card needed. Claude, GPT and Gemini models are paid: add the $5 starter pack at inferall.ai/billing and it becomes spendable balance.