Who Codex-X is for#
Developers managing multiple Codex provider accounts
If you maintain separate official Codex accounts for different clients or projects, or route some work through third-party providers to control costs, Codex-X lets you store each configuration as a named entry and switch between them without touching TOML files. The connection test and model fetch functions confirm a provider is working before you commit to switching.
Skip if:
If you use only a single OpenAI account and never switch providers, the management overhead Codex-X solves does not apply to your workflow. Direct TOML editing is sufficient.
Developers maintaining a personal prompt library
If you maintain custom prompt templates for tasks like code review, debugging, or documentation, Codex-X turns those into a named, categorized library with per-prompt enable/disable controls and two injection modes. You can also pull from 11 bundled and community-synced templates covering software development and writing workflows.
Skip if:
If you use a single static prompt and rarely change it, the prompt center adds little over editing the instruction file directly.
Power users tracking Codex token consumption
Codex-X provides a usage stats view filtered by date and model, showing daily token trends, cache hit rates, and model distribution. Sub-agent usage is attributed to the parent session so you can see the true token cost of each agentic run, not just the top-level session.
Skip if:
If you have a flat-rate Codex subscription with no per-token billing concern, the usage stats are informational but not decision-driving.
Users managing Skills and MCP servers across projects
Instead of navigating the Codex Skills directory and editing MCP configuration files manually, Codex-X surfaces all installed Skills and MCP servers in one screen with individual enable/disable controls. You can install a Skill from a ZIP file, check for updates, and preview existing MCP servers before importing them.
Skip if:
If you run a fixed set of Skills with no need to switch them between sessions, the management UI saves little over the default Codex configuration.
The problem it solves#
When you run OpenAI Codex regularly, the configuration surface grows quickly. Switching from an official OpenAI account to a third-party API provider means opening ~/.codex/config.toml, editing the base URL and API key by hand, restarting Codex, and reversing the change when you switch back. Doing this across multiple projects with separate prompt templates for each workflow turns into a manual maintenance task.
The problem compounds with sessions and Skills. Codex stores conversation history as local files, but there is no built-in way to search, group, or clean them by project. Skills and MCP servers require navigating multiple directories and config files to enable or disable. Token usage stays opaque unless you parse the local database yourself. The result is a tool with strong agentic capabilities and a configuration experience that requires constant file editing.
How it solves it#
Named provider management with one-click switching
Store multiple official Codex accounts and third-party API providers as named configurations in a single view. Each entry includes connection testing, model fetching, and a one-click copy function to duplicate a config as a starting point. Presets for DeepSeek, Qianwen, MiniMax, Kimi, and GLM fill the required fields automatically. Existing cc-switch configurations import directly from the provider setup screen.
Visual prompt injection center
Maintain a categorized prompt library with 5 offline templates and 6 GitHub-synced templates covering software development and writing workflows (11 total). Each prompt has an independent enable/disable toggle and two injection modes: append to existing prompts or replace them entirely. Import your own .md files, add categories, and edit titles and content in the app. Online templates cache locally so they remain available offline.
Session search and project grouping
Search local Codex sessions by title, project path, provider, or session ID. Group sessions by project path to find and continue specific conversations across long histories. Sync a session to the current provider configuration without modifying chat content. Delete sessions individually, by multi-select, or by selecting all sessions under a project. Deletion is permanent and removes the data from Codex's own storage.
Skills and MCP management in one screen
View all installed Skills and MCP servers from a single interface instead of navigating multiple directories. Import existing content, install a Skill from a ZIP file, enable or disable each item individually, and check installed Skills for available updates. Preview existing MCP servers before deciding which ones to bring under Codex-X management.
Token usage statistics by date and model
The settings pane provides a usage tab that filters local token consumption by date range and model, showing daily trends, cache hit rate, model distribution, and the most recent 10 sessions. Sub-agent usage rolls up into its parent session rather than counting as a separate entry, giving you an accurate per-run cost view.
TOML and auth config editor with automatic backup
Read and edit `config.toml` and `auth.json` through the app UI with syntax highlighting. Changes write back to the Codex configuration directory. A backup is created automatically before each significant write. The provider edit screen also exposes the full TOML editor for fine-grained changes, and the 1M context window toggle is available directly from the provider configuration page for supported models.
Strengths and trade-offs#
Strengths
- All data stays on your machineCodex-X stores configuration, login state, and session data locally. No account is required for the app, and no data is uploaded to any Codex-X server. The app accesses remote services only for three specific operations: querying official account quota, testing a third-party connection, and syncing prompt templates from GitHub. Your model requests go directly to whichever provider you configured.
- MIT licensed, no usage restrictionsThe app is MIT licensed and free to download, run, and fork commercially. On Windows and Linux, the portable distribution requires no installation at all. The source is on GitHub with a Rust/Tauri backend that you can inspect before running. Unlike proprietary AI coding tools that require ongoing subscriptions, Codex-X itself has no licensing cost.
- Third-party providers reduce per-session costBy routing requests to third-party models with a Responses-compatible API (DeepSeek, Qianwen, and others), you can lower the per-token cost of Codex sessions without changing your editor workflow. Each provider is stored as a named configuration so switching back to an official OpenAI account takes one click. New or reopened Codex sessions pick up the change immediately, without a full client restart.
- Lightweight Tauri 2 architectureThe app is built with a Rust backend and React/TypeScript frontend via Tauri 2, which produces small native binaries without bundling a Chromium or Electron runtime. Configuration and session data are stored in a local SQLite database at `~/.codexx/codexx.db`, keeping the install footprint minimal.
Trade-offs
- -Requires OpenAI Codex as a prerequisiteCodex-X only works as a companion to OpenAI Codex desktop or CLI. It reads and writes the Codex configuration directory and has no standalone function. Developers using Cursor, Windsurf, or GitHub Copilot as their primary coding assistant get no benefit from it without first adopting Codex.
- -Young project, created July 2026The repository was created in July 2026 with 14 open issues as of September 2026. Edge cases around provider compatibility, session sync behavior, and TOML parsing may appear as the tool matures. Session deletion is permanent with no undo, so the risk of accidental data loss is real for users not paying close attention to the confirmation screen.
- -Per-machine configuration, no syncAll configuration data is stored locally, so there is no built-in way to synchronize Codex-X provider settings or prompt libraries between machines. Developers working across multiple computers must configure providers and prompts on each machine separately.
Codex-X vs alternatives#
Codex-X vs Native Codex Configuration
OpenAI Codex stores all configuration in ~/.codex/config.toml and auth.json. Switching providers, enabling a prompt template, or checking token usage requires direct file edits or parsing the local session database manually. Codex-X wraps those file operations in a GUI and adds capabilities the native interface does not include.
| Feature | Codex-X | Native Codex Config |
|---|---|---|
| License | MIT | Proprietary (OpenAI) |
| Provider switching | Named configs, one click | Edit config.toml |
| Prompt templates | 11 templates, visual toggle | Edit instruction file |
| Session management | Search, group, bulk delete | File system only |
| Token usage view | Daily trends by model | Not available |
| Cost | Free | Requires Codex subscription |
Codex-X is the better choice for anyone running multiple provider configurations or maintaining a reusable prompt library. The native Codex path is sufficient if you use one fixed configuration and rarely change it.
Codex-X vs Cursor
Cursor is a paid AI IDE with built-in support for multiple AI models and a unified context window that integrates directly with the editor. Unlike Codex-X, Cursor is a full editor replacement; it does not sit alongside a separate coding environment.
| Feature | Codex-X | Cursor |
|---|---|---|
| License | MIT | Proprietary |
| Self-hosting | Desktop app, local data | Cloud-connected |
| Scope | Codex companion | Full IDE with AI |
| Pricing | Free | $20/mo (Pro) |
| Multi-provider | Third-party APIs via config | Built-in multi-model |
Cursor is worth considering if you want a single editor that integrates AI without managing separate configurations. Codex-X is the better fit if you already rely on OpenAI Codex for agentic workflows and want better control over prompts, providers, and session history without switching editors.
Quick start#
Install Codex-X by building from source with pnpm and the Tauri 2 toolchain.
```bash
git clone https://github.com/yynxxxxx/Codex-X
cd Codex-X
pnpm install
pnpm dev
```What it's built on#
- Languages
- JavaScriptPythonRustTypeScript
- Frameworks
- React
FAQ#
Does Codex-X require a separate account or subscription?
No. Codex-X is free to download and requires no account of its own. You bring your existing OpenAI Codex credentials (official account or third-party API keys); the app manages those for you locally. There is no Codex-X server that handles your model requests.
Is it safe to store my API keys in Codex-X?
All API keys and login credentials are stored in your machine's local Codex configuration directory and Codex-X's local SQLite database. The app accesses remote services only for three operations: querying your official account quota, testing a third-party connection endpoint, and syncing prompt templates from GitHub. Your model requests go directly to whichever provider you configured, not through any Codex-X proxy.
Can I use DeepSeek or other third-party models with Codex through Codex-X?
Yes, for providers that offer a Responses-compatible API endpoint. Codex-X includes presets for DeepSeek, Qianwen, MiniMax, Kimi, and GLM. After selecting a preset, enter the API key and save. Open a new Codex session or reopen an existing one; existing sessions may need to re-select their model. Actual model availability depends on your account and plan with that provider.
How is Codex-X different from cc-switch?
cc-switch is a command-line provider switcher for Codex. Codex-X covers the same provider-switching capability and adds a GUI, plus prompt management, session search, Skills/MCP management, and token usage stats. Codex-X can import an existing cc-switch configuration directly from the provider setup screen, so you do not need to re-enter credentials.
What happens if I delete a session in Codex-X?
Session deletion is permanent. Codex-X removes the conversation from Codex's own storage, including any derived sub-sessions. The confirmation screen shows the list of sessions about to be deleted. There is no undo. The README explicitly warns: close any Codex windows using those sessions before deleting, and review the confirmation list carefully.
Similar open-source tools#
rakazo
AI teammates you own: your keys, your model, your machine.
autoclip
AI highlight extraction from long videos, no cloud upload
OpenSpec
Spec-driven development for AI coding assistants
tradingview-mcp
MCP bridge for AI-assisted TradingView chart analysis
cli
Official Lark/Feishu CLI with 200+ commands and AI Agent Skills
Multica
Assign GitHub issues to AI coding agents. They run on your machine.