AITabby/codexsplit

What it solves

CodexSplit transforms Codex Desktop into a local AI workstation by allowing users to integrate third-party models, manage multiple official GPT accounts, and route sub-agent tasks, all while maintaining the native functionality of Codex Desktop (such as Computer Use and MCP).

How it works

It uses a Native Gateway + SQLite Trigger Dual-Routing Architecture. Third-party models are routed via native HTTP requests to a local gateway (127.0.0.1:8765), bypassing the need for process interception for standard use. For users who want to rotate multiple official GPT accounts to avoid rate limits, a Desktop Bridge acts as a stdio interceptor to dynamically inject credentials.

Who it’s for

Users of Codex Desktop who want to use models from various providers (like Claude, Gemini, DeepSeek, Qwen) or manage a pool of official GPT accounts to bypass usage quotas without losing native Codex features.

Highlights

• Multi-Provider Support: Easy integration of API Key providers and local OAuth subscriptions (e.g., Grok, Cursor, Claude).
• Official Account Pooling: Quota-weighted round-robin and failover for official GPT accounts to handle 429 errors.
• Agent Routing: Ability to define "model capability profiles" to automatically assign sub-tasks to the most suitable model.
• Session Management: Tools to scan and import local sessions from other AI agents (e.g., Claude Code, Cursor Agent).
• Voice Integration: Configurable STT/TTS settings and a dedicated voice bar for AI interactions.