DevDiff

Appearance

AI Providers Overview ​

DevDiff supports multiple AI backends. You bring your own — DevDiff is just the orchestrator.

Provider Comparison ​

ProviderCostPrivacySpeedQualityOffline
Ollama (Local)Free✅ 100% localMediumGood–Great✅ Yes
WebGPU (Browser)Free✅ 100% localFast (GPU)Good✅ Yes
OpenAIPay-per-use⚠️ Data sent to OpenAIFastExcellent❌ No
AnthropicPay-per-use⚠️ Data sent to AnthropicFastExcellent❌ No
CustomVariesDependsVariesVariesDepends

Default Provider: Ollama ​

Ollama is the default and recommended provider. It runs entirely on your machine — no API keys, no data leaving your system, no costs.

bash
# Install Ollama + pull a model
ollama pull llama3.2:3b

# DevDiff auto-detects it — no config needed
devdiff generate

→ Full Ollama Setup Guide

Configuration ​

All providers are configured in .devdiff.config.js:

javascript
export default {
  ai: {
    providers: [
      // DevDiff tries providers in priority order (lowest number first)
      { name: "local-ollama", url: "ollama://llama3.2:3b", priority: 1 },
      { name: "openai-fallback", url: "openai://gpt-4o-mini", priority: 2 },
    ],
  },
};

If the first provider fails (e.g., Ollama not running), DevDiff automatically tries the next one.

Privacy Model ​

┌─────────────────────────────────────────────────────┐
│  LOCAL (100% Private)                               │
│  • Ollama — models run on your CPU/GPU              │
│  • WebGPU — models run in your browser GPU         │
│  • No data transmitted. No logs. No telemetry.     │
├─────────────────────────────────────────────────────┤
│  CLOUD (Data leaves your machine)                   │
│  • OpenAI — sent to OpenAI API                     │
│  • Anthropic — sent to Anthropic API               │
│  • Subject to their privacy policies               │
└─────────────────────────────────────────────────────┘

DevDiff's data classification engine can automatically strip secrets, PII, and sensitive patterns before they're sent to any cloud provider. See Security Model.

Choosing a Provider ​

Use Ollama if:

  • Privacy is your top priority
  • You don't want API costs
  • You want offline capability
  • Your machine has 8GB+ RAM

Use OpenAI/Anthropic if:

  • You need the highest quality analysis
  • You're reviewing very complex security-sensitive diffs
  • Your machine is underpowered for local models

Use WebGPU if:

  • You're running DevDiff in a browser context
  • You have a capable GPU and want maximum speed

Provider Guides ​