Codex++ — A Management Layer for the Codex Desktop App

Codex++ (project name: BigPizzaV3/CodexPlusPlus) is a companion management tool for the official OpenAI Codex / ChatGPT desktop application. It adds a "++" layer on top: unified control over AI providers and models, better chat sessions, and workspace customization.

What It Does

You open the Codex++ configuration window, set things up, then click Launch to start the official Codex app with your settings applied. Codex++ works through the Chrome DevTools Protocol (CDP) and its own built-in runtime services. It does not modify the official app's app.asar file or write patch files into the install directory.

Key Capabilities

Provider & Model Management

• Four usage modes: official login, official login + API (requests always go through your configured API), pure API (no official account needed), and aggregate providers.
• Aggregate providers come in two types: failover (switch between providers on failure, with rotation/weighting) and model aggregation (route each model to a specific member provider). Members can be API providers or logged-in official accounts.
• Per-model settings: set a custom context window and auto-compression threshold for each model (e.g., deepseek-v4-pro[1M]).
• Model picker groups by provider with collapse/expand, showing provider name and model count.
• Native Fast unlock: lets pure-API-key users use Codex's native Standard / ⚡ Fast controls on supported models (off by default; actual speed and billing depend on the provider).

Session Management

• Scan, delete (with undo), and batch-manage local sessions.
• Export sessions to Markdown, view token history, move projects, and see thread IDs.

Request Reliability

• Request shock absorption: automatically retries when the upstream provider returns overload errors (server_overloaded, at capacity, etc.), so brief congestion doesn't kill your session. Two modes: streaming (keeps real-time output, hands overload errors to Codex's native retry after first token) and full-buffer (buffers the whole response, retries any overload — but loses streaming).
• Request status line shows current stage, model, elapsed time, retry count, and last failure reason.

Workspace Enhancements

• Usage status bar under the input box: real-time generation speed, cache hit rate, and context usage (data from local Codex session records).
• Input & reading comforts: voice input, paste fixing, session width, scroll-position restore, next-step suggestions, answer outlines, and cursor typing effects (rainbow particles, fireworks, starlight).
• Extensions: browse and install community user scripts (a separate market from Codex's official plugin store), plus MCP, Skill, and Codex plugin management.
• Usage widget & desktop pet (dev preview): token stats, task status, draggable/resizable, with replaceable character images and skins.

Multi-Instance & Collaboration

• Codex multi-launch: start isolated Codex instances per provider identity, each with its own config, sessions, and user-data directory. Running instances are listed centrally and can be restarted or stopped individually.
• Sub-agent collaboration: in a Codex conversation, delegate tasks in natural language to Codex CLI, DeepSeek Harness (dsh), or Claude Code; results return to the current conversation. You can set a default agent, model routing, and reuse DSH's saved login.
• WeChat connection: scan a QR code to link local Codex, with each contact mapped to an independent session (configurable access).

Platforms & Tech

• Supported: Windows (x64) and macOS (Apple Silicon / Intel, universal DMG).
• Built with: Rust + Tauri.
• License: AGPL-3.0-only.

Quick Start

1. Download the installer for your platform from GitHub Releases.
2. Make sure the official Codex / ChatGPT desktop app is already installed.
3. Open Codex++, confirm the official app path, choose a provider mode, enable enhancements as desired, and save.
4. Click Launch in the Codex++ UI to start the official app with your configuration.

All UI enhancements can be toggled off individually, or via a master switch; provider and launch management still work with enhancements off.

Where Data Lives

Data	Location
Codex config & login	~/.codex/config.toml, ~/.codex/auth.json
Local database	~/.codex/sqlite/*.db (older fallback: ~/.codex/state_5.sqlite)
Codex++ state & logs	~/.codex-session-delete/
Provider sync backups	~/.codex/backups_state/provider-sync

(If CODEX_HOME is set, that takes precedence over ~/.codex.)

Development Layout

• Frontend: apps/codex-plus-manager
• Core Rust library: crates/codex-plus-core
• Data layer: crates/codex-plus-data
• UI injection scripts: assets/inject

Notes & Caveats

• The project is sponsored by several API relay/proxy services (JOJO Code, APIKEY.FUN, GrooRoute, RunAPI, FennoAI, 百可为AI, 七牛云, TokenMy, APINEBULA). Their claims about service scope, pricing, and promotions are theirs, not the project's.
• Codex++ depends on the official app's page structure, CDP, and local data formats; after official app updates, some features may need adaptation.
• The whale-widget feature is adapted from the upstream project MeteorNOX/DeepSeek-Balance-Whale-Widget; its assets follow their own licenses.
• The usage widget and desktop pet are in development preview; screenshots come from dev builds.
• API keys are stored locally on your machine; the README advises hiding keys before submitting feedback.