Codex Bridge for Copilot Chat

Use OpenAI Codex models directly from the GitHub Copilot Chat model picker in Visual Studio Code with your ChatGPT Plus or Pro subscription.

This extension is a native VS Code LanguageModelChatProvider. It handles OpenAI OAuth locally, streams ChatGPT Codex responses into Copilot Chat, and keeps tokens in VS Code Secret Storage.

Features

• ChatGPT OAuth with PKCE and automatic access-token refresh
• Models discovered from the live Codex catalog for the signed-in ChatGPT account
• Streaming responses, reasoning summaries, images, and tool calling
• Per-model reasoning effort, opt-in Web Search, and Speed Mode controls
• Native Copilot context-window accounting from Codex inference token usage
• Automatic OpenAI prompt-cache reuse for eligible prefixes across normal chat turns and agent tool loops
• Status-bar indicator for five-hour/weekly ChatGPT Codex quota and locally tracked tokens
• Browser callback, manual callback, and optional Codex CLI session import
• Connection test and privacy-safe diagnostics commands

The model picker follows the live Codex catalog instead of a list bundled with the extension. Display names, descriptions, context windows, image/tool capabilities, reasoning efforts, and Fast availability are derived from the signed-in account's catalog response. Models that advertise the fast additional speed tier expose a native Speed Mode control on their normal model entry. Every model also exposes an opt-in Web Search control beside its reasoning controls. It is disabled by default; when enabled, the request includes the Codex Responses API's hosted web_search tool and preserves streamed search annotations as provider data.

Context limits follow Codex's own runtime policy: the catalog's effective-window percentage defines the usable hard window, while its auto-compaction threshold defines the input budget. The remaining headroom is reported to VS Code as the output budget so Copilot Chat displays the same total context limit instead of adding an unrelated output allowance.

Requirements

• VS Code 1.125 or newer
• GitHub Copilot Chat installed and enabled
• A ChatGPT account with Codex access (typically Plus or Pro)

Quick start

1. Install Codex Bridge for Copilot Chat. You need VS Code 1.125 or newer, GitHub Copilot Chat, and a ChatGPT account with Codex access.
2. Run Codex Bridge: Manage Connection from the Command Palette.
3. Choose Sign in with ChatGPT and complete the browser flow.
4. In Copilot Chat, open the model picker, choose Manage Models, enable Codex Bridge, and select a Codex model.

Model names and descriptions come from the live catalog, with dashes in display labels formatted as spaces. Each model exposes native Copilot Chat controls for the advertised reasoning effort and, when supported, Speed Mode, so you can switch to the Fast service tier without adding a second model entry.

If another process uses local port 1455, choose Sign in manually. If the Codex CLI is already signed in, Import Codex CLI Session can copy its OAuth session from ~/.codex/auth.json into VS Code Secret Storage.

Commands

Settings

• openaiCodex.reasoningSummary: workspace default for reasoning summaries; choose auto, concise, detailed, or none, while model follows the live catalog default
• Legacy openaiCodex.speedMode and openaiCodex.reasoningEffort settings remain supported as workspace fallbacks for existing configurations. Prefer the live model picker for new selections.

Each model defaults to the catalog's default_reasoning_level and exposes only the reasoning efforts advertised by that catalog. Fast processing can be selected through the model's native Speed Mode control; it uses more account capacity. Models that accept reasoning.summary also expose a per-model summary control; that selection overrides the workspace default. Web Search is an independent per-model toggle and remains disabled unless selected in the model picker.

• openaiCodex.requestTimeoutSeconds: total request timeout
• openaiCodex.debugLogging: request metadata only; prompts and tokens are never logged
• openaiCodex.showUsageStatusBar: show or hide the Codex quota/token indicator

The status bar reads the same ChatGPT Codex quota endpoint used by Codex CLI and shows the primary and secondary utilization windows. Click it for reset times, available banked reset credits when the account exposes them, the last inference token count, and cumulative tokens tracked by this extension on the current device. A banked reset applies to both the 5-hour and weekly windows; redeeming one requires explicit confirmation.

Documentation

• Setup, settings, and troubleshooting
• OAuth and security
• Development and releases

Related projects

• Grok for GitHub Copilot Chat — Use xAI Grok models directly from the GitHub Copilot Chat model picker in Visual Studio Code.
• Poolside for GitHub Copilot Chat — Use hosted Poolside coding models directly from the GitHub Copilot Chat model picker in Visual Studio Code.

Privacy and status

OAuth tokens are stored through VS Code Secret Storage and are sent only to OpenAI authentication and ChatGPT Codex endpoints. This is an independent community extension, not an official OpenAI or GitHub product. The ChatGPT Codex backend is not a public compatibility API and can change; see docs/security.md.

Requests identify this extension as openai-oauth-copilot-chat; they do not claim to originate from the official Codex CLI or VS Code extension. The shared OAuth client and ChatGPT backend are undocumented integration surfaces, so OpenAI may change or revoke compatibility without notice.

Unofficial project; not affiliated with OpenAI, GitHub, or Microsoft. The ChatGPT Codex backend is not a public compatibility API and can change. Licensed under MIT.