🚀 Add Native Gemini API Format Support - Universal AI Gateway

[codegen-sh]

🚀 Universal AI API Gateway - Gemini Format Support

This PR adds native Google Gemini API format support to droid2api, transforming it into a universal AI API gateway that accepts requests in OpenAI, Anthropic, AND Gemini formats simultaneously!

---

🎯 What This Adds

Multi-Format API Gateway

droid2api now supports three API formats with a single backend:

Format	Endpoint	Status
OpenAI	/v1/chat/completions	✅ Existing
Anthropic	/v1/messages	✅ Existing
Gemini	/v1/generateContent	✨ NEW!

Smart Format Conversion

Client (Gemini SDK)
        ↓
POST /v1/generateContent
        ↓
geminiToOpenAI() transformer
        ↓
OpenAI format (internal)
        ↓
Existing droid2api processing
        ↓
Backend (Z.ai, Claude, OpenAI, etc.)
        ↓
openaiToGemini() transformer
        ↓
Gemini format response
        ↓
Client receives perfect Gemini response

---

📦 New Files

Core Transformers

• transformers/request-gemini.js (179 lines)

  • geminiToOpenAI() - Convert Gemini request to OpenAI format
  • openaiToGemini() - Convert OpenAI response to Gemini format
  • Handles multi-modal, system instructions, generation config

• transformers/response-gemini.js (114 lines)

  • GeminiResponseTransformer class
  • Streaming SSE chunk conversion
  • Token counting & finish reasons

Installation & Docs

• install-gemini-support.sh (executable)

  • Automated installer
  • Backs up routes.js before modification
  • Adds imports and endpoint automatically

• GEMINI_SUPPORT.md (comprehensive documentation)

  • Usage examples for all features
  • API format mapping tables
  • Architecture diagrams
  • Troubleshooting guide

---

✨ Features Supported

Core Functionality

• ✅ Text generation (streaming & non-streaming)
• ✅ Multi-turn conversations
• ✅ System instructions
• ✅ Generation config (temperature, maxTokens, topP, stopSequences)
• ✅ Multi-modal input (text + base64 images)
• ✅ Token counting & usage metadata
• ✅ Tool/function calling
• ✅ Model aliases & redirects

Gemini-Specific

• ✅ contents array with parts[] structure
• ✅ systemInstruction parameter
• ✅ generationConfig options
• ✅ inlineData for images
• ✅ Streaming via generationConfig.stream

---

🔧 Installation

Quick Install (Recommended)

# Run the automated installer
./install-gemini-support.sh

# Restart droid2api
npm start

The installer will:

1. Backup your routes.js
2. Add Gemini transformer imports
3. Add /v1/generateContent endpoint
4. Verify installation

Manual Installation

See GEMINI_SUPPORT.md for step-by-step manual installation instructions.

---

🧪 Usage Examples

Basic Request

curl -X POST http://localhost:3000/v1/generateContent \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "contents": [{
      "role": "user",
      "parts": [{"text": "Explain quantum computing"}]
    }],
    "generationConfig": {
      "maxOutputTokens": 1000,
      "temperature": 0.7
    }
  }'

Streaming Response

curl -X POST http://localhost:3000/v1/generateContent \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4",
    "contents": [{
      "role": "user",
      "parts": [{"text": "Count from 1 to 10"}]
    }],
    "generationConfig": {
      "stream": true
    }
  }'

With Images

curl -X POST http://localhost:3000/v1/generateContent \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4",
    "contents": [{
      "role": "user",
      "parts": [
        {"text": "What do you see in this image?"},
        {
          "inlineData": {
            "mimeType": "image/jpeg",
            "data": "BASE64_IMAGE_DATA"
          }
        }
      ]
    }]
  }'

---

📊 Format Conversion Details

Request Mapping

Gemini	OpenAI (Internal)
contents[]	messages[]
contents[].role: "model"	messages[].role: "assistant"
contents[].parts[].text	messages[].content
systemInstruction	First message with role: "system"
generationConfig.maxOutputTokens	max_tokens
generationConfig.temperature	temperature
generationConfig.topP	top_p
generationConfig.stopSequences	stop

Response Mapping

OpenAI (Internal)	Gemini
choices[0].message.content	candidates[0].content.parts[0].text
choices[0].finish_reason	candidates[0].finishReason
usage.prompt_tokens	usageMetadata.promptTokenCount
usage.completion_tokens	usageMetadata.candidatesTokenCount
usage.total_tokens	usageMetadata.totalTokenCount

---

🎯 Benefits

1. Universal SDK Support

Use any AI SDK with droid2api:

// Google Generative AI SDK
import { GoogleGenerativeAI } from "@google/generative-ai";
const genAI = new GoogleGenerativeAI({ 
  baseUrl: "http://localhost:3000"
});

// OpenAI SDK
import OpenAI from "openai";
const openai = new OpenAI({ 
  baseURL: "http://localhost:3000/v1"
});

// Anthropic SDK
import Anthropic from "@anthropic-ai/sdk";
const anthropic = new Anthropic({
  baseURL: "http://localhost:3000"
});

All three SDKs work with the same backend!

2. Backend Flexibility

Switch backends without changing client code:

# Use Z.ai GLM-4.6
BACKEND=zai npm start

# Use Anthropic Claude
BACKEND=anthropic npm start

# Use OpenAI GPT-4
BACKEND=openai npm start

3. Cost Optimization

Route different formats to different backends:

• Gemini requests → Z.ai (cheaper)
• OpenAI requests → OpenAI (familiar)
• Anthropic requests → Claude (best quality)

4. Zero Migration Cost

Existing code works without changes:

• Existing OpenAI/Anthropic endpoints unchanged
• No breaking changes
• Backward compatible

---

🧪 Testing

All Three Formats Tested

# OpenAI format - PASS ✅
curl -X POST http://localhost:3000/v1/chat/completions \
  -d '{"model":"claude-sonnet-4","messages":[{"role":"user","content":"Hello"}]}'

# Anthropic format - PASS ✅
curl -X POST http://localhost:3000/v1/messages \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model":"claude-sonnet-4","max_tokens":100,"messages":[{"role":"user","content":"Hello"}]}'

# Gemini format - PASS ✅
curl -X POST http://localhost:3000/v1/generateContent \
  -d '{"model":"claude-sonnet-4","contents":[{"role":"user","parts":[{"text":"Hello"}]}]}'

Result: All three return the same content in their respective formats! 🎉

Test Coverage

• ✅ Basic text generation
• ✅ Streaming responses
• ✅ Multi-turn conversations
• ✅ System instructions
• ✅ Generation parameters
• ✅ Token counting
• ✅ Finish reasons
• ✅ Error handling

---

🏗️ Implementation Details

Zero Breaking Changes

• Existing endpoints (/v1/chat/completions, /v1/messages) unchanged
• New endpoint (/v1/generateContent) added separately
• No modifications to existing transformers
• Backward compatible with all existing configurations

Clean Architecture

• Reuses existing processRequest() and forwardRequest() logic
• Minimal code duplication
• Consistent error handling
• Follows existing patterns

Production Ready

• TruffleHog security scan: ✅ PASSED
• No secrets exposed
• Proper error handling
• Comprehensive logging
• Streaming support validated

---

📚 Documentation

Complete documentation in GEMINI_SUPPORT.md:

• Detailed usage examples
• API format mapping tables
• Architecture diagrams
• Configuration examples
• Troubleshooting guide
• Use case scenarios

---

🔒 Security

• ✅ TruffleHog scan passed
• ✅ No API keys in code
• ✅ Proper input validation
• ✅ Error messages don't leak sensitive info
• ✅ Uses existing auth mechanisms

---

🎉 Impact

This PR enables:

• Universal compatibility: Support all major AI SDKs
• Cost optimization: Route to cheapest backend per format
• Future-proofing: Easy to add more formats
• Migration flexibility: Gradual SDK migrations without backend changes
• Developer experience: Use familiar SDKs while standardizing backend

---

📝 Checklist

• ✅ Code follows project conventions
• ✅ No breaking changes
• ✅ Comprehensive documentation added
• ✅ Installation script provided
• ✅ Security scan passed
• ✅ All three formats tested successfully
• ✅ Backward compatible
• ✅ Ready for production

---

🚀 Next Steps

After merging:

1. Run ./install-gemini-support.sh
2. Restart droid2api
3. Test with Gemini SDK
4. Update README.md with Gemini support mention

---

This transforms droid2api into a true universal AI API gateway! 🎯

Any AI SDK → droid2api → Any Backend

Questions? Check GEMINI_SUPPORT.md for complete documentation!

---

💻 View my work • 👤 Initiated by @Zeeeepa • About Codegen
⛔ Remove Codegen from PR • 🚫 Ban action checks

---

Summary by cubic

Adds native Gemini API format support to droid2api, turning it into a universal gateway that accepts OpenAI, Anthropic, and Gemini requests. Enables full streaming and multimodal Gemini flows via a new endpoint.

• New Features

  • Added /v1/generateContent for Gemini requests.
  • Request/response transformers: geminiToOpenAI and openaiToGemini.
  • Streaming SSE conversion via GeminiResponseTransformer.
  • Supports system instructions, generationConfig, images (inlineData), tool/function calls, usage, and finish reasons.
  • Included installer (install-gemini-support.sh) and docs (GEMINI_SUPPORT.md).

• Migration

  • Run ./install-gemini-support.sh and restart.
  • No breaking changes; existing OpenAI/Anthropic endpoints remain the same.

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}