DevDiff

Appearance

OpenAI Setup ​

Use OpenAI's GPT models with DevDiff for the highest-quality changelog analysis.

Privacy Note: When using OpenAI, your diff content is sent to OpenAI's servers. See Privacy Guarantees for how DevDiff's data classification engine protects sensitive data before transmission.

Prerequisites ​

  1. An OpenAI account
  2. A paid OpenAI API plan (free tier has rate limits)
  3. An API key from platform.openai.com/api-keys

Setup ​

1. Get Your API Key ​

  1. Go to platform.openai.com/api-keys
  2. Click Create new secret key
  3. Copy the key — it starts with sk-

2. Set the Environment Variable ​

bash
# Add to your shell profile (.zshrc, .bashrc, etc.)
export OPENAI_API_KEY="sk-your-key-here"

# Or create a .env file in your project (DevDiff reads this automatically)
echo 'OPENAI_API_KEY=sk-your-key-here' >> .env

Never commit your API key to Git. Add .env to your .gitignore.

3. Configure DevDiff ​

javascript
// .devdiff.config.js
export default {
  ai: {
    providers: [
      {
        name: "openai-gpt4o",
        url: "openai://gpt-4o-mini", // Model to use
        priority: 1, // Try this first
      },
    ],
  },
};

4. Generate Changelog ​

bash
devdiff generate

Available Models ​

ModelCostSpeedQualityBest For
gpt-4o-mini~$0.01/analysisFastGreatDaily use, most diffs
gpt-4o~$0.10/analysisFastExcellentComplex security reviews
gpt-4-turbo~$0.15/analysisMediumExcellentVery large diffs

Costs are estimates for a typical 200-line diff. Actual costs depend on token usage.

javascript
// Use a specific model
{
  url: "openai://gpt-4o-mini";
}
{
  url: "openai://gpt-4o";
}
{
  url: "openai://gpt-4-turbo";
}

Using OpenAI as a Fallback ​

The most common setup: use Ollama locally (free, private) and fall back to OpenAI when needed:

javascript
// .devdiff.config.js
export default {
  ai: {
    providers: [
      { name: "local", url: "ollama://llama3.2:3b", priority: 1 },
      { name: "cloud-fallback", url: "openai://gpt-4o-mini", priority: 2 },
    ],
  },
};

DevDiff tries Ollama first. If it's not running, it falls back to OpenAI automatically.

Cost Control ​

javascript
// .devdiff.config.js — limit token usage
export default {
  ai: {
    providers: [
      {
        name: "openai",
        url: "openai://gpt-4o-mini",
        maxTokens: 1000, // Limit output length
        priority: 1,
      },
    ],
  },
};

Troubleshooting ​

"Invalid API key"

bash
# Verify your key is set
echo $OPENAI_API_KEY

# Test directly
curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"

"Insufficient quota"

"Rate limit exceeded"

  • DevDiff auto-retries with backoff
  • Add Ollama as a local fallback provider