Bring Your Own AI (BYOAI)

DevDiff's core philosophy: You control the AI, not us.

Provider Tiers

Local & Free (No API Keys)

Provider	Setup	Best For
Ollama	ollama pull [model]	macOS, Linux, Windows WSL
llama.cpp	Point to binary path	Custom builds, embedded systems
Transformers.js	Zero config	Node.js without native deps
WebLLM	Zero config	Browser-based, Chrome

Self-Hosted (Your Servers)

Provider	Config Key
LocalAI	localai://localhost:8080
vLLM	vllm://localhost:8000
Text Generation Inference	tgi://localhost:3000
Ollama (remote)	ollama://my-server:11434

Cloud (Your API Keys)

Provider	Config Key
OpenAI	openai://gpt-4o-mini
Anthropic	anthropic://claude-3-haiku
Google Gemini	gemini://gemini-1.5-flash
Groq	groq://llama3-70b
DeepSeek	deepseek://deepseek-coder
Together AI	together://meta-llama/Llama-3-70b
Fireworks AI	fireworks://accounts/fireworks/models/llama-v3-70b

Smart Routing

// .devdiff.config.js
export default {
  ai: {
    providers: [
      {
        name: "local-small",
        url: "ollama://llama3.2:3b",
        maxTokens: 2000,
        priority: 1, // Try first
      },
      {
        name: "local-big",
        url: "ollama://llama3:70b",
        maxTokens: 8000,
        priority: 2, // Complex diffs
      },
      {
        name: "cloud-cheap",
        url: "openai://gpt-4o-mini",
        apiKey: process.env.OPENAI_API_KEY,
        maxTokens: 4000,
        priority: 3, // Fallback
        maxDailyCost: 0.5, // Budget cap!
      },
    ],

    routing: {
      strategy: "cost-aware", // 'priority' | 'cost-aware' | 'latency'
      complexityThreshold: 0.6,
      localOnly: false,
    },
  },
};

Privacy Guarantees

When using local providers:

• ✅ Zero network requests for AI processing
• ✅ All diffs processed on-device
• ✅ No telemetry, no analytics, no tracking
• ✅ Audit log shows every AI call

When using cloud providers:

• ✅ Only structured diffs sent (not raw files)
• ✅ Automatic secret redaction
• ✅ Dry-run mode shows exactly what's sent
• ✅ .devdiffignore excludes sensitive paths