# Provider Integration Content for Landing Page

**Suggested Section**: Add this content to https://ruv.io/agentic-flow as a new "Provider Support" section

---

## 🔌 Multi-Provider LLM Support

**agentic-flow** works with multiple LLM providers through intelligent proxy architecture that automatically converts requests to provider-specific formats while maintaining full compatibility with Claude Agent SDK.

### Supported Providers

| Provider | Models | Cost vs Claude | Speed | Setup Required |
|----------|--------|----------------|-------|----------------|
| **Anthropic** | Claude 3.5 Sonnet, Opus, Haiku | Baseline | Fast | `ANTHROPIC_API_KEY` |
| **Gemini** (NEW v1.9.3) | Gemini 2.0 Flash, Pro | **70% cheaper** | **2-5x faster** | `GOOGLE_GEMINI_API_KEY` |
| **OpenRouter** | 100+ models (GPT-4, Llama, DeepSeek) | Varies | Varies | `OPENROUTER_API_KEY` |
| **ONNX** | Phi-4 (runs locally) | **100% FREE** | Medium | No key needed |

### Quick Provider Examples

```bash
# Anthropic (default, highest quality)
npx agentic-flow --agent coder --task "Build REST API"

# Gemini (fastest, cheapest - NEW in v1.9.3!)
npx agentic-flow --agent coder --task "Build REST API" --provider gemini

# OpenRouter (access 100+ models)
npx agentic-flow --agent coder --task "Build REST API" --provider openrouter

# ONNX (100% free, runs locally)
npx agentic-flow --agent coder --task "Build REST API" --provider onnx
```

### Gemini Provider - Fully Functional (v1.9.3)

The **Gemini provider is now production-ready** with complete streaming support. Three critical bugs were fixed in v1.9.3:

✅ **Model Selection** - Correctly uses Gemini models instead of Claude model names
✅ **Streaming Responses** - Full Server-Sent Events (SSE) support with `&alt=sse` parameter
✅ **Provider Selection** - Respects `--provider` flag and doesn't auto-select incorrectly

**Gemini Benefits:**
- ⚡ **2-5x faster response times** compared to Anthropic
- 💰 **70% cost reduction** - great for high-volume tasks
- 🎯 **Excellent for** code generation, code analysis, refactoring, simple tasks
- ✅ **Full streaming support** for real-time responses
- 🔄 **Zero code changes** - drop-in replacement via proxy architecture

**Setup:**
```bash
# Get your free Gemini API key from https://ai.google.dev/
export GOOGLE_GEMINI_API_KEY=your_key_here

# Use Gemini with streaming
npx agentic-flow --agent coder --task "Create function" --provider gemini --stream
```

### How Provider Proxies Work

**Intelligent Request Conversion:**

1. **User makes request** → Claude Agent SDK format (Anthropic Messages API)
2. **Proxy intercepts** → Converts to provider-specific format
3. **Provider responds** → Native format (Gemini, OpenRouter, etc.)
4. **Proxy converts back** → Claude Agent SDK format
5. **User receives** → Standard response, no code changes needed

**Architecture Benefits:**
- ✅ **Single codebase** works with all providers
- ✅ **Provider-agnostic** agent development
- ✅ **Easy switching** between providers via CLI flags
- ✅ **Cost optimization** - use cheap providers for simple tasks, Claude for complex ones
- ✅ **Automatic failover** - switch providers if one is down

### Configuration

**Environment Variables:**
```bash
# Provider API Keys (set only what you need)
export ANTHROPIC_API_KEY=sk-ant-...        # Anthropic Claude
export GOOGLE_GEMINI_API_KEY=AIza...       # Google Gemini
export OPENROUTER_API_KEY=sk-or-...        # OpenRouter

# Model Selection (optional)
export COMPLETION_MODEL=claude-sonnet-4-5-20250929  # Anthropic model
export REASONING_MODEL=gemini-2.0-flash-exp         # Gemini model
```

**CLI Flags:**
```bash
--provider <name>    # anthropic, gemini, openrouter, onnx
--model <model-id>   # Override default model
--stream             # Enable streaming responses
```

### Cost Optimization Strategy

**Smart Provider Selection:**

```bash
# Simple tasks → Use Gemini (70% cheaper, 5x faster)
npx agentic-flow --agent coder --task "Add comments" --provider gemini

# Complex reasoning → Use Claude (highest quality)
npx agentic-flow --agent researcher --task "Analyze architecture" --provider anthropic

# Experimentation → Use ONNX (100% free)
npx agentic-flow --agent tester --task "Generate tests" --provider onnx

# Access specific models → Use OpenRouter
npx agentic-flow --agent coder --task "Optimize code" --provider openrouter --model meta-llama/llama-3.2-3b-instruct
```

### Production Deployment

**High-Volume Applications:**
```bash
# Use Gemini for 80% of requests (fast, cheap)
export DEFAULT_PROVIDER=gemini
export GOOGLE_GEMINI_API_KEY=...

# Fall back to Claude for complex tasks
export ANTHROPIC_API_KEY=...

# Run with automatic provider selection
npx agentic-flow --agent coder --task "..." # Uses Gemini by default
npx agentic-flow --agent architect --task "..." --provider anthropic # Override for complex tasks
```

**Monitoring:**
- Track token usage per provider
- Monitor response quality
- Measure cost savings
- Analyze performance metrics

---

## Why This Matters for Consulting Clients

**Cost Savings:**
- **70% reduction** in LLM costs using Gemini for appropriate tasks
- **100% free** local inference with ONNX for development/testing
- **Flexible budgeting** - scale up/down providers based on needs

**Performance:**
- **2-5x faster** responses with Gemini for code generation
- **Real-time streaming** for better user experience
- **Local inference** option eliminates API latency

**Flexibility:**
- **100+ models** available via OpenRouter
- **Zero vendor lock-in** - switch providers anytime
- **Multi-provider strategies** - use best tool for each job

---

## Technical Specifications

**Proxy Architecture:**
- Located in `src/proxy/` directory
- `anthropic-to-gemini.ts` - Gemini proxy (v1.9.3 - fully functional)
- `anthropic-to-openrouter.ts` - OpenRouter proxy
- Automatic request/response format conversion
- Full MCP tool support via function calling conversion
- Streaming and non-streaming modes

**Supported Features:**
- ✅ Text generation
- ✅ Streaming responses (SSE)
- ✅ Tool calling / Function calling
- ✅ System prompts
- ✅ Multi-turn conversations
- ✅ Temperature control
- ✅ Token limits

**Verified Compatibility:**
- Claude Agent SDK
- Claude Code CLI
- MCP (Model Context Protocol) tools
- All 66 specialized agents
- Multi-agent swarm orchestration

---

## Get Started

```bash
# Install agentic-flow
npm install -g agentic-flow@1.9.3

# Setup your preferred provider
export GOOGLE_GEMINI_API_KEY=your_key_here

# Run your first agent with Gemini
npx agentic-flow --agent coder --task "Write a hello world function" --provider gemini --stream

# See the cost and speed difference yourself!
```

**Documentation:** https://github.com/ruvnet/agentic-flow#readme
**Full Provider Guide:** https://github.com/ruvnet/agentic-flow#-provider-support
**Latest Release:** v1.9.3 - Gemini Provider Fully Functional