InferenceIndex

Early access · Private preview

Getting started

Using with coding agents

Point Cursor, Claude Code, and Codex at the OpenAI-compatible base URL.

The API is in early access, a private preview. Use the key from the dashboard, one base URL, and one model name.

Cursor, Claude Code, and Codex pointed at one base URL
SettingValue
Base URLhttps://api.inferenceindex.com/v1
Modelworkflow/auto
API keyThe key from the dashboard at https://inferenceindex-app.web.app

Leave /chat/completions off the base URL. The client adds that path.

Cursor

  1. Open Cursor Settings, then Models.
  2. Paste the dashboard key into the OpenAI API key field.
  3. Turn on Override OpenAI Base URL and set https://api.inferenceindex.com/v1.
  4. Add the model name workflow/auto and select it.

Cursor sends OpenAI-compatible chat completions to that base URL for local Chat and Agent requests that use this model.

Claude Code

Claude Code's own base URL, ANTHROPIC_BASE_URL, speaks the Anthropic Messages API. Leave that unset for this host. https://api.inferenceindex.com/v1 is the OpenAI-compatible base URL.

For OpenAI-compatible clients in the project Claude Code is editing, set:

bash
export OPENAI_BASE_URL="https://api.inferenceindex.com/v1"
export OPENAI_API_KEY="YOUR_KEY"

Use the model workflow/auto in the client, as in the quickstart. You can paste this into the session:

text
Call the Compiled Agent Model at base URL https://api.inferenceindex.com/v1
with model workflow/auto. The API is OpenAI-compatible chat completions.
Read the key from the environment. Do not print the key.

Codex

Codex reads the user file ~/.codex/config.toml. A project-local config file cannot change the provider. Put the base URL in the user file:

toml
model = "workflow/auto"
model_provider = "inferenceindex"

[model_providers.inferenceindex]
name = "InferenceIndex"
base_url = "https://api.inferenceindex.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"

Put the dashboard key in OPENAI_API_KEY. Current Codex sends OpenAI Responses requests to that base URL plus /responses. The early-access call documented in the quickstart is chat completions: POST /v1/chat/completions with model workflow/auto. Use the quickstart from code when you want that call directly.