[Feature] Add Codex as Model Provider with OpenAI OAuth Integration

[Clawiee]

Pre-checks

• I have searched existing issues and this is not a duplicate.

Feature Request

Add Codex as a Model Provider with OpenAI OAuth Integration

Problem Statement

Currently, Clawith does not support Codex as a model provider. Users who want to use OpenAI's Codex for code-related tasks must configure API keys manually, which can be inconvenient and less secure.

Proposed Solution

Add a new model provider for OpenAI Codex with the following features:

1. OAuth-based Authentication: Users can sign in with their GPT/OpenAI account via OAuth instead of manually entering API keys
2. Quota Management: Users can use their own Codex quota/tokens from their GPT subscription
3. Secure Authentication: Leverage OpenAI's official OAuth flow for secure authentication
4. User-Friendly Setup: Simple "Sign in with OpenAI" button to connect

Use Cases

• Code generation and review
• Code interpretation and execution
• Developer-focused AI assistants
• Teams using Codex-enabled GPT accounts

Benefits

• Better UX: No need to manage API keys manually
• Cost Efficiency: Users can use existing Codex quota from their GPT subscription
• Security: OAuth is more secure than sharing API keys
• Broader Accessibility: Makes Codex more accessible to non-technical users

Additional Context

Reference: OpenAI OAuth documentation for integrating GPT/Codex access.

---

Feature request submitted by Yifei Zhou

[Clawiee]

Analysis: Adding Codex as Model Provider with OpenAI OAuth

Thanks for this feature request, @Clawiee! This is an interesting enhancement that would allow users to leverage their GPT/Codex subscription quota through OAuth authentication.

Technical Feasibility Assessment

Based on the current codebase analysis, here's what would be needed:

1. Model Provider Integration (backend/app/models/llm.py)

The LLMModel table already supports a provider field (currently stores values like "anthropic", "openai", "deepseek"). A new provider type "codex" would be straightforward to add.

2. OpenAI OAuth Implementation (backend/app/services/auth_provider.py)

The existing BaseAuthProvider framework provides a solid foundation. A new OpenAIAuthProvider class would follow the same pattern as existing providers:

class OpenAIAuthProvider(BaseAuthProvider):
    provider_type = "openai_oauth"
    
    OPENAI_TOKEN_URL = "https://oauth.ai.google.com/developerconsole/v1/exchange"
    OPENAI_USER_INFO_URL = "https://api.openai.com/v1/me"
    # Note: OpenAI OAuth for GPT/Codex access requires specific setup

3. Key Implementation Components

• OAuth Flow: Sign in with OpenAI → Exchange code for access token → Store encrypted token per user
• Token Storage: Similar to Google Workspace OAuth in PR feat(gws): Google Workspace integration with OAuth, tools, skill discovery & scope management #326 - AES-256-CBC encryption
• API Integration: Use OpenAI's API with OAuth tokens instead of API keys
• Quota Management: OpenAI's usage API can be queried to track user quotas

4. UI Components Needed

• "Connect with OpenAI" button on model settings page
• OAuth callback handler
• Token status indicator (connected/disconnected)

Important Considerations

1. OpenAI OAuth Scope: OpenAI's consumer OAuth is primarily for their ChatGPT features. For Codex access, you may need to check if OpenAI exposes Codex via the same OAuth flow or requires separate API key management.

2. Reference Implementation: PR feat(gws): Google Workspace integration with OAuth, tools, skill discovery & scope management #326 (Google Workspace OAuth) provides an excellent reference for the token management architecture.

3. Security: All OAuth tokens should be encrypted at rest (AES-256-CBC as used elsewhere in the codebase).

Suggested Implementation Approach

Phase	Description
1	Create OpenAIAuthProvider class in auth_provider.py
2	Add token storage model for per-user OAuth tokens
3	Implement OAuth callback endpoint
4	Update LLM client to use OAuth tokens for Codex models
5	Frontend UI for OAuth connection flow

Questions for Clarification

1. Should Codex OAuth tokens be per-user (each user connects their own OpenAI account) or per-tenant (one org-wide OAuth connection)?
2. Do you have specific Codex models in mind (e.g., codex-code-001, codex-davinci-002)?

---

Labels: This aligns with the existing enhancement. No additional labels needed.

Priority: Medium - Nice to have for user experience improvement, but requires careful OAuth implementation.

[Y1fe1Zh0u]

@codex review this issue

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.