AITabby/codexsplit

What it solves

CodexSplit acts as a local control center for Codex Desktop, allowing users to integrate third-party AI models, manage multiple official GPT accounts, and configure advanced agent routing without losing the native functionality of the Codex application.

How it works

It operates as a local gateway (typically on port 8765) that sits between Codex Desktop and various AI providers. It manages API keys, OAuth subscriptions, and model metadata, then exposes these models to Codex Desktop via a "Desktop Bridge." It maintains four distinct routing paths: native official GPT conversations, native GPT-Live interactions, third-party model conversations via the gateway, and specific child-task handoffs (spawn_agent) that use a custom Agent routing system.

Who it’s for

Power users of Codex Desktop who want to use models from different providers (like DeepSeek, Qwen, or Kimi) alongside official GPT models, or those who need to manage multiple official accounts and complex agent-based workflows on their local machine.

Highlights

• Multi-Provider Gateway: Supports presets for various providers and custom OpenAI-compatible endpoints with API key pooling (round-robin or failover).
• Agent Routing: Allows users to create "Agent Profiles" with specific expertise descriptions to automatically route sub-tasks to the most suitable model.
• Official Account Pool: Manages multiple official ChatGPT/Codex accounts with weighted round-robin scheduling based on remaining quota.
• Session Management: Ability to scan and import local sessions from other AI agents like Cursor, Claude Code, and Grok CLI.
• Voice Integration: Configurable STT/TTS settings and a dedicated voice bar for AI interactions.