YOUR CODEX, UPGRADED — OPEN SOURCE · LOCAL · RUST + TAURI

Make Codex nicer++

Manage providers and models in one place, make conversations easier to use,
and shape your workspace around you. No app.asar patches, nothing written into install dirs.

Download v1.8.0 GitHub

0: GitHub Stars
0: Mainstream models
0: Max per-model context
0: Files patched in the official app

LIVE DEMO · TYPING EFFECTS

Every keystroke answers back

Not a video — the real particle engine shipped in Codex++, running on your screen right now. Type something.

Codex composer · codex++

＋ Attach ⛨ Full access DeepSeek Flash · Max ▾ ↑

FEATURES · THE MATRIX

One manager, top to bottom

Every enhancement has its own switch; with the master switch off, provider and launch management keep working.

Switch providers freely

Official sign-in, official + API, API-only, and aggregates. Failover switches members on errors; model aggregation pins models to members. Responses / Chat Completions conversion, model tests and Provider Doctor included.

Every model, its own space

Set context windows and compaction thresholds per model. Codex++ generates a dedicated model_catalog_json so Codex reads each model's window. Import model metadata; pick MCP, skills and plugins per provider.

Simpler session management

Local session scanning, delete with undo, batch management, Markdown export, token history, project moves and thread IDs.

Extend your workspace

Browse community user scripts for model selectors, usage stats and more. Manage MCP servers, skills and Codex plugins, or hot-reload your own scripts.

Smoother writing and reading

Voice input, paste fixes, conversation width, scroll restoration, next-step suggestions, answer outlines — and the typing effects you just tried.

Many Codex, one machine

Launch isolated Codex instances per provider identity: separate config, sessions and data directories, running side by side, each restartable on its own.

Usage status line

Live generation speed, cache hit rate and context usage below the composer, read from local Codex session records with matching definitions. On by default.

Usage, with company

Usage widgets and desktop pets: token stats, task status and budget alerts. Drag, resize, reskin, or bring your own character art.

WeChat connect

Scan a QR code to reach your local Codex. Each contact maps to an isolated session, with an allow-list of WeChat users. Hand off work from anywhere.

Delegate in a sentence

Ask your current Codex chat to delegate to DeepSeek Harness (dsh), Claude Code or another Codex. The selected CLI runs the task and returns results to the same chat. Configure the default agent, CLI paths and models; naming an agent selects that target.

Native Fast with an API key

Restore the native Fast switch in API-only mode and use Codex's own Standard / ⚡ Fast controls. Off by default; the model must declare Fast (priority) support. Availability and billing depend on your provider.

Request shock absorption

Absorb model capacity/overload errors (server_overloaded, server_is_overloaded, at capacity) at the proxy with backoff retries so Codex sessions are interrupted less often. The status line shows the request phase, model, elapsed time, retry count and last failure reason; streaming mode keeps live output while buffered mode absorbs overloads at any point.

PROVIDER MODES

Four ways in, your call

Each mode is configured independently and switches in one click — the current profile is saved before the target one is written.

Official sign-in

Use your ChatGPT / Codex account. Requests go through the official channel with the full official experience and plugin entry.

Sign in with your ChatGPT / Codex account
Official quota and plugin marketplace
Zero config, works out of the box

You Codex OpenAI

SHOWCASE · REAL UI

What you see is what ships

Real Codex++ UI and feature graphics. Select a thumbnail to preview; click the large image to view it full size.

A Codex chat asks DeepSeek what 1+1 is, uses Codex Plus Collaboration, then shows DeepSeek's answer: 1+1 equals 2 — **Sub-agents, in action**Codex delegates to DeepSeek via dsh and returns the result in the same chat. Click to view full size.

Codex++ Manager: models, sessions, input behavior and widget settings in one place — **Enhancements hub**Plugins & models, dialogs & input, widgets & pets — one page of switches.

Codex multi-launch: running DeepSeek and Zhipu GLM instances — **Multi-launch**One isolated instance per provider identity, no cross-talk.

Model picker folded by provider: Provider 4 group (Default, 7 models) and DeepSeek group (2 models) — **Grouped model picker**Folded by provider, with counts and a Default badge.

Codex++ extension marketplace: community scripts and workflow tools — **Extension market**Install community scripts in one click: usage panels, model matrix, prompt tools.

Request status and automatic retry: send a request, wait while the model is busy, then resume generation after a successful retry — **Request status and automatic retry**Wait and retry automatically when the model is busy, with progress and the last failure reason shown inline.

Request log and automatic retry: see in-flight requests, model capacity errors, retry counts and successful recovery in real time — **Request log and automatic retry**See in-flight requests in real time and follow every retry when a model is at capacity.

Usage status line: 51 tok/s, 99.9% cache hits, 19% context — **Usage status line**51 tok/s · 99.9% cache hits · 19% context, right under the composer.

Desktop pet usage widget with token stats bubble — **Pet widget**Today's usage, task status and budget alerts — swap in your own character.

Rainbow particle typing effects in the Codex composer — **Typing effects**Rainbow particles, fireworks and starlight; pause automatically with Reduce Motion.

PER-MODEL CONTEXT

One suffix, a million tokens

Write a window suffix in the model list and Codex++ generates a dedicated model_catalog_json so Codex uses the right window per model. Models without a suffix keep provider-level settings.

model_list · config

# Write window suffixes right in the model list
deepseek-v4-pro[1M]
claude-opus-5.5[200K]
your-model[128000]

# Codex++ generates model_catalog_json
# codex reads each model's own window ↓

✓ catalog injected config.toml · model_catalog_json

generic-model

200K

claude-opus-5.5

400K

deepseek-v4-pro

Each model declares its own window; the status line's context share follows the same per-model numbers.

GETTING STARTED

Three steps, done

Codex++ works alongside the official Codex / ChatGPT desktop app — install that first.

Grab the installer

From GitHub Releases: Windows picks *-windows-x64-setup.exe, macOS (Apple Silicon / Intel) picks *-macos-universal.dmg — drag it into Applications.

Open Codex++ and configure

Double-click always opens the manager: confirm the official app path, pick a provider mode, toggle enhancements, save. API keys stay on your machine.

Launch Codex from the manager

Hit "Launch" to load your saved provider and enhancement config into the official app. Injection-based settings apply after a Codex restart.

Every enhancement can be toggled individually; with the master switch off, provider and launch management keep working.

CHANGELOG

Always moving

Highlights from the latest releases — see the repo CHANGELOG and Releases for the full log.

FAQ

Asked, answered

Why is there no Codex++ menu after launch?

Make sure you launched the official app from the Codex++ manager with the "Launch" button. Check the app path, launch status and diagnostic logs in the manager's Maintenance and About pages.

Why do requests fail after switching providers?

Run a model test or Provider Doctor in the provider details first: verify protocol, Base URL, key and model. API-only and official+API modes store credentials differently — don't hand-copy auth.json between them.

Why is the native Fast switch still missing?

Confirm that you saved and restarted Codex through Codex++. In your provider's model settings, set “Fast (priority) support”, then save both the model and provider before restarting. Only declare support that your provider actually offers; unlocking the client control does not change service limits or billing.

Can enhancements be disabled? Will they survive official app updates?

Yes — individually or via the master switch. Codex++ depends on the official app's page structure, CDP and local data formats, so some features may need follow-up adaptations after official updates. Back up before changing provider settings or session data.

Where are config, sessions and logs stored?

CODEX_HOME wins when set; otherwise ~/.codex. API keys never leave your machine — redact secrets before submitting feedback.

COMMUNITY & SUPPORT

Built together

Open an issue with your OS, Codex++ version, repro steps and redacted logs. Extensions, themes and tips are always welcome.

GitHub IssuesBugs & feature requests↗ QQ Group 41127858981↗ Telegram@CodexPlusPlus↗ LINUX DOCommunity forum↗

Your Codex deserves these two pluses

Free & open source · AGPL-3.0 · Windows and macOS
Your config and sessions never leave your machine.

Windows · x64 *-windows-x64-setup.exe Download → macOS · Universal *-macos-universal.dmg Download →

Works with the official Codex / ChatGPT desktop app. Open Codex++, configure a provider, and hit "Launch" in the manager.