Inference Gateway

The Inference Gateway is a proxy server designed to facilitate access to various language model APIs. It allows users to interact with different language models through a unified interface, simplifying the configuration and the process of sending requests and receiving responses from multiple LLMs, enabling an easy use of Mixture of Experts.

• Key Features
• Overview
• Installation
• Middleware Control and Bypass Mechanisms
• Model Context Protocol (MCP) Integration
• Metrics and Observability
• Supported API's
• Configuration
• Examples
• SDKs
• CLI Tool
• Contributing
• License

• 📜 Open Source: Available under the MIT License.
• 🚀 Unified API Access: Proxy requests to multiple language model APIs, including OpenAI, Ollama, Ollama Cloud, Groq, Cohere etc.
• ⚙️ Environment Configuration: Easily configure API keys and URLs through environment variables.
• 🔧 Tool-use Support: Enable function calling capabilities across supported providers with a unified API.
• 🌐 MCP Support: Full Model Context Protocol integration - automatically discover and expose tools from MCP servers to LLMs without client-side tool management.
• 🌊 Streaming Responses: Stream tokens in real-time as they're generated from language models.
• 🖼️ Vision/Multimodal Support: Process images alongside text with vision-capable models.
• 🐳 Docker Support: Use Docker and Docker Compose for easy setup and deployment.
• ☸️ Kubernetes Support: Ready for deployment in Kubernetes environments.
• 📊 OpenTelemetry: Monitor and analyze performance.
• 🛡️ Production Ready: Built with production in mind, with configurable timeouts and TLS support.
• 🌿 Lightweight: Includes only essential libraries and runtime, resulting in smaller size binary of ~10.8MB.
• 📉 Minimal Resource Consumption: Designed to consume minimal resources and have a lower footprint.
• 📚 Documentation: Well documented with examples and guides.
• 🧪 Tested: Extensively tested with unit tests and integration tests.
• 🛠️ Maintained: Actively maintained and developed.
• 📈 Scalable: Easily scalable and can be used in a distributed environment with HPA in Kubernetes.
• 🔒 Compliance and Data Privacy: This project does not collect data or analytics, ensuring compliance and data privacy.
• 🏠 Self-Hosted: Can be self-hosted for complete control over the deployment environment.
• ⌨️ CLI Tool: Improved command-line interface for managing and interacting with the Inference Gateway

You can horizontally scale the Inference Gateway to handle multiple requests from clients. The Inference Gateway will forward the requests to the respective provider and return the response to the client.

Note: MCP middleware components can be easily toggled on/off via environment variables (MCP_ENABLE) or bypassed per-request using headers (X-MCP-Bypass), giving you full control over which capabilities are active.

Note: Vision/multimodal support is disabled by default for security and performance. To enable image processing with vision-capable models (GPT-4o, Claude 4.5, Gemini 2.5, etc.), set ENABLE_VISION=true in your environment configuration.

The following diagram illustrates the flow:

%%{init: {'theme': 'base', 'themeVariables': { 'primaryColor': '#326CE5', 'primaryTextColor': '#fff', 'lineColor': '#5D8AA8', 'secondaryColor': '#006100' }, 'fontFamily': 'Arial', 'flowchart': {'nodeSpacing': 50, 'rankSpacing': 70, 'padding': 15}}}%%


graph TD
    %% Client nodes
    A["👥 Clients / 🤖 Agents"] --> |POST /v1/chat/completions| Auth

    %% Auth node
    Auth["🔒 Optional OIDC"] --> |Auth?| IG1
    Auth --> |Auth?| IG2
    Auth --> |Auth?| IG3

    %% Gateway nodes
    IG1["🖥️ Inference Gateway"] --> P
    IG2["🖥️ Inference Gateway"] --> P
    IG3["🖥️ Inference Gateway"] --> P

    %% Middleware Processing and Direct Routing
    P["🔌 Proxy Gateway"] --> MCP["🌐 MCP Middleware"]
    P --> |"Direct routing bypassing middleware"| Direct["🔌 Direct Providers"]
    MCP --> |"Middleware chain complete"| Providers["🤖 LLM Providers"]

    %% MCP Tool Servers
    MCP --> MCP1["📁 File System Server"]
    MCP --> MCP2["🔍 Search Server"]
    MCP --> MCP3["🌐 Web Server"]

    %% LLM Providers (Middleware Enhanced)
    Providers --> C1["🦙 Ollama"]
    Providers --> D1["🚀 Groq"]
    Providers --> E1["☁️ OpenAI"]

    %% Direct Providers (Bypass Middleware)
    Direct --> C["🦙 Ollama"]
    Direct --> D["🚀 Groq"]
    Direct --> E["☁️ OpenAI"]
    Direct --> G["⚡ Cloudflare"]
    Direct --> H1["💬 Cohere"]
    Direct --> H2["🧠 Anthropic"]
    Direct --> H3["🐋 DeepSeek"]

    %% Define styles
    classDef client fill:#9370DB,stroke:#333,stroke-width:1px,color:white;
    classDef auth fill:#F5A800,stroke:#333,stroke-width:1px,color:black;
    classDef gateway fill:#326CE5,stroke:#fff,stroke-width:1px,color:white;
    classDef provider fill:#32CD32,stroke:#333,stroke-width:1px,color:white;
    classDef mcp fill:#FF69B4,stroke:#333,stroke-width:1px,color:white;

    %% Apply styles
    class A client;
    class Auth auth;
    class IG1,IG2,IG3,P gateway;
    class C,D,E,G,H1,H2,H3,C1,D1,E1,Providers provider;
    class MCP,MCP1,MCP2,MCP3 mcp;
    class Direct direct;

Client is sending:

curl -X POST http://localhost:8080/v1/chat/completions
  -d '{
    "model": "openai/gpt-3.5-turbo",
    "messages": [
      {
        "role": "system",
        "content": "You are a pirate."
      },
      {
        "role": "user",
        "content": "Hello, world! How are you doing today?"
      }
    ],
  }'

** Internally the request is proxied to OpenAI, the Inference Gateway inferring the provider by the model name.

You can also send the request explicitly using ?provider=openai or any other supported provider in the URL.

Finally client receives:

{
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "Ahoy, matey! 🏴‍☠️ The seas be wild, the sun be bright, and this here pirate be ready to conquer the day! What be yer business, landlubber? 🦜",
        "role": "assistant"
      }
    }
  ],
  "created": 1741821109,
  "id": "chatcmpl-dc24995a-7a6e-4d95-9ab3-279ed82080bb",
  "model": "N/A",
  "object": "chat.completion",
  "usage": {
    "completion_tokens": 0,
    "prompt_tokens": 0,
    "total_tokens": 0
  }
}

For streaming the tokens simply add to the request body stream: true.

Recommended: For production deployments, running the Inference Gateway as a container is recommended. This provides better isolation, easier updates, and simplified configuration management. See Docker or Kubernetes deployment examples.

The Inference Gateway can also be installed as a standalone binary using the provided install script or by downloading pre-built binaries from GitHub releases.

The easiest way to install the Inference Gateway is using the automated install script:

Install latest version:

curl -fsSL https://raw.githubusercontent.com/inference-gateway/inference-gateway/main/install.sh | bash

Install specific version:

curl -fsSL https://raw.githubusercontent.com/inference-gateway/inference-gateway/main/install.sh | VERSION=v0.22.3 bash

Install to custom directory:

# Install to custom location
curl -fsSL https://raw.githubusercontent.com/inference-gateway/inference-gateway/main/install.sh | INSTALL_DIR=~/.local/bin bash

# Install to current directory
curl -fsSL https://raw.githubusercontent.com/inference-gateway/inference-gateway/main/install.sh | INSTALL_DIR=. bash

What the script does:

• Automatically detects your operating system (Linux/macOS) and architecture (x86_64/arm64/armv7)
• Downloads the appropriate binary from GitHub releases
• Extracts and installs to /usr/local/bin (or custom directory)
• Verifies the installation

Supported platforms:

• Linux: x86_64, arm64, armv7
• macOS (Darwin): x86_64 (Intel), arm64 (Apple Silicon)

Download pre-built binaries directly from the releases page:

1. Download the appropriate archive for your platform

2. Extract the binary:

   tar -xzf inference-gateway_<OS>_<ARCH>.tar.gz

3. Move to a directory in your PATH:

   sudo mv inference-gateway /usr/local/bin/
   chmod +x /usr/local/bin/inference-gateway

inference-gateway --version

Once installed, start the gateway with your configuration:

# Set required environment variables
export OPENAI_API_KEY="your-api-key"

# Start the gateway
inference-gateway

For detailed configuration options, see the Configuration section below.

The Inference Gateway uses middleware to process requests and add capabilities like MCP (Model Context Protocol). Clients can control which middlewares are active using bypass headers:

• X-MCP-Bypass: Skip MCP middleware processing

# Use only standard tool calls (skip MCP)
curl -X POST http://localhost:8080/v1/chat/completions \
  -H "X-MCP-Bypass: true" \
  -d '{
    "model": "anthropic/claude-3-haiku",
    "messages": [{"role": "user", "content": "Connect to external agents"}]
  }'

# Skip both middlewares for direct provider access
curl -X POST http://localhost:8080/v1/chat/completions \
  -H "X-MCP-Bypass: true" \
  -d '{
    "model": "groq/llama-3-8b",
    "messages": [{"role": "user", "content": "Simple chat without tools"}]
  }'

For Performance:

• Skip middleware processing when you don't need tool capabilities
• Reduce latency for simple chat interactions

For Selective Features:

• Use only standard tool calls (skip MCP): Add X-MCP-Bypass: true
• Direct provider access

For Development:

• Test middleware behavior in isolation
• Debug tool integration issues
• Ensure backward compatibility with existing applications

The middlewares use these same headers to prevent infinite loops during their operation:

MCP Processing:

• When tools are detected in a response, the MCP agent makes up to 10 follow-up requests
• Each follow-up request includes X-MCP-Bypass: true to skip middleware re-processing
• This allows the agent to iterate without creating circular calls

Note: These bypass headers only affect middleware processing. The core chat completions functionality remains available regardless of header values.

Enable MCP to automatically provide tools to LLMs without requiring clients to manage them:

# Enable MCP and connect to tool servers
export MCP_ENABLE=true
export MCP_SERVERS="http://filesystem-server:3001/mcp,http://search-server:3002/mcp"

# LLMs will automatically discover and use available tools
curl -X POST http://localhost:8080/v1/chat/completions \
  -d '{
    "model": "openai/gpt-4",
    "messages": [{"role": "user", "content": "List files in the current directory"}]
  }'

The gateway automatically injects available tools into requests and handles tool execution, making external capabilities seamlessly available to any LLM.

Learn more: Model Context Protocol Documentation | MCP Integration Example

The Inference Gateway provides comprehensive OpenTelemetry metrics for monitoring performance, usage, and function/tool call activity. Metrics are automatically exported to Prometheus format and available on port 9464 by default.

# Enable telemetry and set metrics port (default: 9464)
export TELEMETRY_ENABLE=true
export TELEMETRY_METRICS_PORT=9464

# Access metrics endpoint
curl http://localhost:9464/metrics

Track token consumption across different providers and models:

• llm_usage_prompt_tokens_total - Counter for prompt tokens consumed
• llm_usage_completion_tokens_total - Counter for completion tokens generated
• llm_usage_total_tokens_total - Counter for total token usage

Labels: provider, model

# Total tokens used by OpenAI models in the last hour
sum(increase(llm_usage_total_tokens_total{provider="openai"}[1h])) by (model)

Monitor API performance and reliability:

• llm_requests_total - Counter for total requests processed
• llm_responses_total - Counter for responses by HTTP status code
• llm_request_duration - Histogram for end-to-end request duration (milliseconds)

Labels: provider, request_method, request_path, status_code (responses only)

# 95th percentile request latency by provider
histogram_quantile(0.95, sum(rate(llm_request_duration_bucket{provider=~"openai|anthropic"}[5m])) by (provider, le))

# Error rate percentage by provider
100 * sum(rate(llm_responses_total{status_code!~"2.."}[5m])) by (provider) / sum(rate(llm_responses_total[5m])) by (provider)

Comprehensive tracking of tool executions for MCP, and standard function calls:

• llm_tool_calls_total - Counter for total function/tool calls executed
• llm_tool_calls_success_total - Counter for successful tool executions
• llm_tool_calls_failure_total - Counter for failed tool executions
• llm_tool_call_duration - Histogram for tool execution duration (milliseconds)

Labels: provider, model, tool_type, tool_name, error_type (failures only)

Tool Types:

• mcp - Model Context Protocol tools (prefix: mcp_)
• standard_tool_use - Other function calls

# Tool call success rate by type
100 * sum(rate(llm_tool_calls_success_total[5m])) by (tool_type) / sum(rate(llm_tool_calls_total[5m])) by (tool_type)

# Average tool execution time by provider
sum(rate(llm_tool_call_duration_sum[5m])) by (provider) / sum(rate(llm_tool_call_duration_count[5m])) by (provider)

# Most frequently used tools
topk(10, sum(increase(llm_tool_calls_total[1h])) by (tool_name))

Complete monitoring stack with Grafana dashboards:

cd examples/docker-compose/monitoring/
cp .env.example .env  # Configure your API keys
docker compose up -d

# Access Grafana at http://localhost:3000 (admin/admin)

Production-ready monitoring with Prometheus Operator:

cd examples/kubernetes/monitoring/
task deploy-infrastructure
task deploy-inference-gateway

# Access via port-forward or ingress
kubectl port-forward svc/grafana-service 3000:3000

The included Grafana dashboard provides:

• Real-time Metrics: 5-second refresh rate for immediate feedback
• Tool Call Analytics: Success rates, duration analysis, and failure tracking
• Provider Comparison: Performance metrics across all supported providers
• Usage Insights: Token consumption patterns and cost analysis
• Error Monitoring: Failed requests and tool call error classification

Learn more: Docker Compose Monitoring | Kubernetes Monitoring | OpenTelemetry Documentation

• OpenAI
• Ollama
• Ollama Cloud (Preview)
• Groq
• Cloudflare
• Cohere
• Anthropic
• DeepSeek
• Google
• Mistral

The Inference Gateway can be configured using environment variables. The following environment variables are supported.

To enable vision capabilities for processing images alongside text:

ENABLE_VISION=true

Supported Providers with Vision:

• OpenAI (GPT-4o, GPT-5, GPT-4.1, GPT-4 Turbo)
• Anthropic (Claude 3, Claude 4, Claude 4.5 Sonnet, Claude 4.5 Haiku)
• Google (Gemini 2.5)
• Cohere (Command A Vision, Aya Vision)
• Ollama (LLaVA, Llama 4, Llama 3.2 Vision)
• Groq (vision models)
• Mistral (Pixtral)

Note: Vision support is disabled by default for performance and security reasons. When disabled, requests with image content will be rejected even if the model supports vision.

• Using Docker Compose
  • Basic setup - Simple configuration with a single provider
  • MCP Integration - Model Context Protocol with multiple tool servers
  • Hybrid deployment - Multiple providers (cloud + local)
  • Authentication - OIDC authentication setup
  • Tools - Tool integration examples
• Using Kubernetes
  • Basic setup - Simple Kubernetes deployment
  • MCP Integration - Model Context Protocol in Kubernetes
  • Agent deployment - Standalone agent deployment
  • Hybrid deployment - Multiple providers in Kubernetes
  • Authentication - OIDC authentication in Kubernetes
  • Monitoring - Observability and monitoring setup
  • TLS setup - TLS/SSL configuration
• Using standard REST endpoints

More SDKs could be generated using the OpenAPI specification. The following SDKs are currently available:

• Typescript
• Rust
• Go
• Python

The Inference Gateway CLI provides a powerful command-line interface for managing and interacting with the Inference Gateway. It offers tools for configuration, monitoring, and management of inference services.

• Status Monitoring: Check gateway health and resource usage
• Interactive Chat: Chat with models using an interactive interface
• Configuration Management: Manage gateway settings via YAML config
• Project Initialization: Set up local project configurations
• Tool Execution: LLMs can execute whitelisted commands and tools

go install github.com/inference-gateway/cli@latest

curl -fsSL https://raw.githubusercontent.com/inference-gateway/cli/main/install.sh | bash

Download the latest release from the releases page.

1. Initialize project configuration:

   infer init

2. Check gateway status:

   infer status

3. Start an interactive chat:

   infer chat

For more details, see the CLI documentation.

This project is licensed under the MIT License.

Found a bug, missing provider, or have a feature in mind?
You're more than welcome to submit pull requests or open issues for any fixes, improvements, or new ideas!

Please read the CONTRIBUTING.md for more details.

My motivation is to build AI Agents without being tied to a single vendor. By avoiding vendor lock-in and supporting self-hosted LLMs from a single interface, organizations gain both portability and data privacy. You can choose to consume LLMs from a cloud provider or run them entirely offline with Ollama.