Magpie

Use this guide when you run Magpie — the menu-bar app and local gateway that points coding agents at one model list — and want that list to include Felo API Platform models. Magpie runs a gateway on 127.0.0.1:3425 and rewrites each agent's configuration to call it, so agents never hold the Felo key or the Felo URL.

Before You Start

You need one Felo API Platform key. The same key works across model protocols, agent clients, Search, Web Fetch, PPT, Mindmap, LiveDocs, X Search, YouTube Subtitling, Landing Page, and SuperAgent APIs.

export FELO_API_KEY="YOUR_API_KEY"

Install Magpie if you have not already:

curl -fsSL https://usemagpie.ai/install.sh | sh

Add Felo as a Provider

Felo serves Chat Completions, Responses, and Anthropic Messages from one key. Give Magpie every base it can use:

magpie provider add felo \
  url=https://openapi.felo.ai/api/v1 \
  responses=https://openapi.felo.ai/api/v1 \
  anthropic=https://openapi.felo.ai/api \
  key="$FELO_API_KEY"

In the app, Providers → Add provider takes the same settings: name felo, the key, and the OpenAI-compatible base https://openapi.felo.ai/api/v1. For an Anthropic-compatible provider, use https://openapi.felo.ai/api.

Magpie asks Felo for the model list once the key is saved; those models show up as felo/<model-id> in every agent's picker.

magpie provider models felo   # re-fetch Felo's list
magpie models                 # what agents can pick, including felo/<model>

Pick a Felo Model in an Agent

magpie claude felo/claude-sonnet-5-5
magpie codex felo/gpt-5.6-sol

Or pick one on the Agents page. Because Felo exposes all three protocols, each agent keeps its native one — Claude Code stays on Anthropic Messages, Codex stays on Responses — and both land on Felo.

Client Fields

Verify the Setup

magpie provider test felo

Magpie sends one small request per protocol and reports latency. Then ask an agent one short question with a felo/ model selected; a reply means agent → Magpie gateway → Felo is working.

Troubleshooting

Next Steps

• Getting Started
• LLM API Reference
• Models and Pricing
• Troubleshooting