> ## Documentation Index
> Fetch the complete documentation index at: https://docs.glyphformac.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Codex ChatGPT OAuth
> Connect to ChatGPT using Codex OAuth authentication
Codex provides OAuth-based access to ChatGPT models, allowing you to use your existing ChatGPT Plus or Team subscription without separate API keys.
## Prerequisites
* Codex CLI installed: [codexlang.com](https://codexlang.com)
* ChatGPT Plus or Team subscription (optional, but provides access to better models)
* Codex must be in your system PATH
## Setup
Install Codex from [codexlang.com](https://codexlang.com):
```bash theme={null}
curl -fsSL https://codexlang.com/install.sh | sh
```
Download the installer from [codexlang.com](https://codexlang.com) and add to PATH.
Verify installation:
```bash theme={null}
codex --version
```
Test that Codex app-server works:
```bash theme={null}
codex app-server
```
You should see JSON-RPC output. Press Ctrl+C to exit.
Go to **Settings → AI** and select the **Codex (ChatGPT OAuth)** profile.
1. Click **Connect** in the Codex account section
2. Your browser opens to ChatGPT OAuth consent page
3. Log in and authorize Codex
4. Return to Glyph
5. Click **Refresh** to verify connection
You should see your email and account status as "connected".
Click the **Model** dropdown. Codex fetches models available to your ChatGPT account.
Available models depend on your subscription:
* **Free tier**: `gpt-3.5-turbo`, `gpt-4o-mini`
* **Plus/Team**: `gpt-4o`, `o1`, `o1-mini`, `o3-mini`
For reasoning models (`o1`, `o3-mini`), select a reasoning effort:
* **low** - Faster responses
* **medium** - Balanced (default)
* **high** - More thorough reasoning
Open the AI panel and send a test message. You should receive a response from ChatGPT.
## How It Works
Glyph runs Codex as a managed child process:
1. When Glyph starts, it launches `codex app-server` via stdio
2. Communication happens over JSON-RPC
3. OAuth tokens stored by Codex (not in Glyph)
4. Glyph sends chat requests through Codex
5. Codex streams responses back to Glyph
## Configuration
### Provider Settings
* **Service**: `codex_chatgpt`
* **Model**: Set automatically from Codex (e.g., `gpt-4o`, `o1-mini`)
* **Authentication**: OAuth via Codex (no API key)
* **Reasoning Effort**: Available for reasoning models
### Account Management
#### Check Connection Status
In Settings → AI → Codex section, you'll see:
* **Status**: `connected` or `disconnected`
* **Email**: Your ChatGPT account email
* **Display Name**: Your ChatGPT account name
* **Auth Mode**: `chatgpt` or similar
#### Refresh Status
Click **Refresh** to update account status and rate limits.
#### Disconnect
Click **Disconnect** to log out of ChatGPT. This clears OAuth tokens stored by Codex.
### Rate Limits
Codex reports ChatGPT rate limits in Glyph settings:
* **Requests per hour**
* **Requests per day**
* **Usage percentage**
* **Reset time**
Rate limits refresh automatically every 30 seconds.
## Reasoning Models
### Available Reasoning Models
* `o1` - OpenAI's original reasoning model
* `o1-mini` - Faster, lighter reasoning model
* `o3-mini` - Latest reasoning model (if available)
### Reasoning Effort Levels
Select reasoning effort in Settings → AI:
| Level | Description | Use Case |
| ---------- | -------------------------------- | --------------------------- |
| **low** | Quick thinking, faster responses | Simple tasks, iterations |
| **medium** | Balanced reasoning (default) | General problem-solving |
| **high** | Deep thinking, slower responses | Complex math, logic puzzles |
Reasoning models show thinking process in the AI panel before providing answers.
## Features
### Chat Mode
Conversational interaction:
* Back-and-forth dialogue
* No file system access
* Uses ChatGPT's web interface rate limits
* Best for Q\&A and brainstorming
### Create Mode
ChatGPT with workspace tools:
* **read\_file** - Read files from your space
* **search\_notes** - Search note content
* **list\_dir** - List directory contents
* Tool usage tracked in timeline view
* Best for research and knowledge retrieval
### Context Attachment
Attach notes for grounded responses:
* Attach files or folders via context menu
* Mention with `@filename` syntax
* Configure character budget (up to 250K chars)
* Context sent through Codex to ChatGPT
## Billing and Costs
Codex uses your existing ChatGPT subscription:
* ✅ No separate API charges
* ✅ No token-by-token billing
* ✅ Flat subscription rate (Plus: $20/month, Team: $25/user/month)
* ✅ Unlimited messages within rate limits
Codex/ChatGPT is ideal if you already have ChatGPT Plus or Team and want to avoid per-token API charges.
## Troubleshooting
### "failed to start codex app-server"
**Cause**: Codex is not installed or not in PATH.
**Solution**:
1. Verify installation: `codex --version`
2. Ensure `codex` is in your system PATH
3. Try running `codex app-server` manually to check for errors
### "codex request timed out"
**Possible causes**:
* Network connectivity issues
* Codex app-server crashed
* ChatGPT API is down
**Solution**:
1. Restart Glyph (this restarts Codex app-server)
2. Check network connectivity
3. Try disconnecting and reconnecting your account
### Login opens browser but status shows disconnected
**Solution**:
1. Click **Refresh** in AI Settings
2. Wait a few seconds for Codex to complete login
3. Check that you authorized in the browser
4. Some OAuth flows complete asynchronously
### "AI responses fail immediately"
**Checklist**:
1. ✅ Profile provider is `Codex (ChatGPT OAuth)`
2. ✅ Model is selected (not empty)
3. ✅ Account status is `connected`
4. ✅ Rate limits not exceeded
**Solution**: Disconnect and reconnect account, or restart Glyph.
### No streaming text appears
**Cause**: Codex app-server may have exited or stopped sending notifications.
**Solution**: Restart Glyph to restart Codex app-server.
### Rate limit exceeded
**Cause**: You've hit ChatGPT's hourly or daily message limits.
**Solution**: Wait for the rate limit to reset (shown in settings). ChatGPT Plus has higher limits than free tier.
### Model list is empty
**Cause**: Codex couldn't fetch models from ChatGPT.
**Solution**:
1. Verify account is connected
2. Click **Refresh**
3. Check that you're logged into ChatGPT in your browser
4. Restart Glyph
## Advanced Configuration
### Codex Logs
Codex app-server logs are managed by Glyph. Check Glyph's logs for Codex communication details.
### Custom Codex Path
If Codex is not in PATH, you'll need to add it. Glyph doesn't support custom Codex paths in the UI.
**macOS/Linux**:
```bash theme={null}
export PATH="$PATH:/path/to/codex"
```
**Windows**:
Add Codex directory to System PATH in Environment Variables.
## Operational Notes
* Glyph runs Codex app-server as a managed child process over stdio JSON-RPC
* OAuth is started via `account/login/start` with `type: "chatgpt"`
* Chat requests use `thread/start` + `turn/start` and stream via notifications
* Authentication tokens stored by Codex, not Glyph
* Codex process lifetime is tied to Glyph (restarts with app)
## Security
* OAuth tokens stored by Codex CLI (not in `.glyph/app/ai_secrets.json`)
* Codex handles authentication flow
* No API keys stored in Glyph
* Communication over local stdio (not network)
## Codex vs OpenAI API
### Use Codex ChatGPT OAuth When:
✅ You have ChatGPT Plus or Team subscription
✅ You want flat-rate billing (no per-token charges)
✅ You prefer OAuth over API keys
✅ You want access to reasoning models included in Plus
### Use OpenAI API When:
✅ You need programmatic access for automation
✅ You want higher rate limits
✅ You need fine-tuned models
✅ You prefer pay-as-you-go billing
## Next Steps
Learn about chat vs create modes
Attach notes to conversations
Compare with direct OpenAI API integration
Codex-specific troubleshooting