LLM API

Claude Code and Codex Configuration

Configure Claude Code and Codex to route model calls through Felo API Platform.

This guide shows how to configure agent clients to use the Felo API Platform LLM API. Use this page when you want Claude Code or Codex to route model calls through Felo API Platform instead of calling a model vendor directly.

Set FELO_API_KEY to your Felo API Platform key before using either client.

export FELO_API_KEY="YOUR_API_KEY"

Supported Client Paths

Client or API surface	Recommended Felo endpoint	Model example	Notes
Claude Code	`POST /api/v1/messages`	`claude-sonnet-4-6`	Claude Code uses the Anthropic Messages wire format.
Codex	`POST /api/v1/responses`	`gpt-5.6-sol`	Codex custom providers should use the Responses API.
Chat Completions	`POST /api/v1/chat/completions`	`gpt-5.6-sol`	Use from SDKs or server-side integrations. Codex Chat Completions support is deprecated in Codex documentation.

Claude Code

Claude Code should use the Anthropic-compatible Messages endpoint.

Claude Code can use any published model listed in Models and Pricing, including the newly available claude-sonnet-5:

claude --model claude-sonnet-5

Shell Configuration

export ANTHROPIC_API_KEY="$FELO_API_KEY"
export ANTHROPIC_BASE_URL="https://openapi.felo.ai/api"
export ANTHROPIC_MODEL="claude-sonnet-4-6"

claude

The Anthropic base URL intentionally omits /v1, because Claude Code and the Anthropic SDK append /v1/messages to the configured API host.

Settings File Configuration

Use a Claude Code settings file when you want the endpoint and model to apply every time claude starts. Keep API keys in your shell or secret manager instead of committing them into project settings.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://openapi.felo.ai/api",
    "ANTHROPIC_MODEL": "claude-sonnet-4-6"
  }
}

You can also choose the model for one session:

claude --model claude-sonnet-4-6

Codex

Codex should use the OpenAI-compatible Responses endpoint. gpt-5.6-sol is the flagship GPT-5.6 tier for complex reasoning, coding, and agentic workflows.

User Config

Add a Felo provider to your user-level Codex configuration at ~/.codex/config.toml.

model_provider = "felo"
model = "gpt-5.6-sol"

[model_providers.felo]
name = "Felo API Platform LLM API"
base_url = "https://openapi.felo.ai/api/v1"
env_key = "FELO_API_KEY"
wire_api = "responses"

Codex endpoint authentication and endpoint selection must live in user-level Codex config. Codex ignores model_provider, model_providers, and openai_base_url from project-scoped .codex/config.toml files.

One-Off Model Selection

After the Felo provider is configured, you can choose the model for a single CLI run:

codex --model gpt-5.6-sol

For non-interactive runs:

codex exec --model gpt-5.6-sol "Review this change"