Home / Code

magpie

Every agent's model. One place. Codex on DeepSeek, Claude Code on Kimi, from the menu bar.

FreeShort listing1 views
Visit website

Sign in to rate this tool.

Overview

magpie screenshot

Magpie provides a single local gateway that lets any programming‑assistant agent use any AI model it needs. Once you add a provider, DeepSeek, Kimi, GLM, Qwen, OpenRouter, Ollama, Claude or ChatGPT, magpie can route that model to agents such as Codex, Claude Code, Gemini CLI, or any agent listed in its menu bar.

The installation is straightforward. A macOS download is available, and it can also be installed from terminal on Intel Macs, Windows, or Linux. After installing, a small daemon runs locally on port 3425. Agents communicate with the gateway through their usual OpenAI or Anthropic APIs; the gateway translates requests and responses, keeping tool calls, streaming, and reasoning intact. Configuration changes are handled on the fly: clicking a model value in the menu bar rewrites only the changed key in the agent's configuration, and the next session automatically uses the new model.

Privacy is maintained by placeholder substitution. When an agent sends a request, magpie intercepts any API keys, tokens, private keys, emails, or card numbers that match its masking rules and replaces them with in‑memory placeholders. The vendor receives only the placeholder, and the placeholder is restored in the reply, both in streamed text and in tool calls. No sensitive values ever leave the user's machine or write to disk.

User accounts from any paid plan are integrated automatically. Signing into an agent with a Claude Pro, ChatGPT Plus, or GitHub Copilot account adds that provider's models to magpie's picker. You can switch profiles such as “Budget” or “Focus” to update all agents at once. Smart routing assigns each turn to the most appropriate model based on intent, ensuring that the right model is used without running out of quota.

Key features

  • One local gateway that supports DeepSeek, Kimi, GLM, Qwen, OpenRouter, Ollama, Claude, and ChatGPT models
  • Real‑time replacement of provider credentials with in‑memory placeholders
  • Automatic routing of requests to any agent that speaks OpenAI or Anthropic APIs
  • Transparent configuration changes from the menu bar
  • Comprehensive masking of secrets, tokens, emails, and card numbers
  • Integration with existing paid account plans for smooth access

Details

Pricing
Free
Category
Code
Suggest an edit

Sign in to suggest an edit.

Pros and cons

Pros

  • Centralizes model access across a wide range of providers
  • Keeps sensitive data on the local machine without disk persistence
  • Offers intuitive control over agent configurations via the menu bar
  • Supports smart routing based on intent for efficient model use

Cons

  • Only available for macOS, Intel Macs, Windows, and Linux
  • Requires local installation and daemon to run
  • No published pricing, community plan, or subscription details provided in the site text

Alternatives