Martian Gateway

Martian Gateway is a product from Martian, an AI research lab, that gives developers a single API endpoint to access over 200 large language models from providers including OpenAI, Anthropic, and Google. It is designed as a drop-in replacement compatible with existing OpenAI and Anthropic SDKs, meaning teams can switch or route between models without rewriting their integration code. The gateway is accompanied by a dashboard for tracking real-time usage, model performance, and request history.

What It Is

Martian Gateway is a unified LLM API gateway — a routing and access layer that sits between your application and multiple AI model providers. Instead of managing separate API keys, SDKs, and endpoint formats for each provider, developers point their requests at a single Martian endpoint (api.withmartian.com/v1/chat/completions) and specify the model by name using a provider/model format. The gateway handles the translation and routing to the underlying provider.

Model Access and Compatibility

The gateway exposes 200+ models and is built to be compatible with the OpenAI and Anthropic API formats, so existing code using those SDKs requires minimal changes. Supported integrations documented by Martian include:

• SDKs: OpenAI SDK, Anthropic SDK, Vercel AI SDK
• AI coding tools: Cursor, Claude Code, Cline, Aider, Codex, OpenCode
• Proxy layers: LiteLLM, HTTP client

This breadth of integration targets makes the gateway usable across both direct API consumers and AI-powered developer tooling.

Open-Source Adapter Library

Alongside the hosted gateway, Martian maintains an open-source Python library called llm-adapters (MIT license, available on GitHub at withmartian/llm-adapters). The library provides a unified interface for calling different LLM providers using the OpenAI SDK format as a common interface, supporting both synchronous and asynchronous operations. As of its latest release (v8.11.0, published May 2025), it requires Python 3.11+ and is installable via pip. The adapter path format follows provider/vendor/model_name, and supported models can be retrieved programmatically via AdapterFactory.get_supported_models().

Where It Fits in the Stack

Martian Gateway is positioned as infrastructure for teams that want model flexibility without vendor lock-in. By abstracting provider differences behind a single API, it allows developers to:

• Swap models without changing application code
• Compare model performance across providers using the dashboard
• Use the same integration pattern across coding assistants, agent frameworks, and direct API calls

The gateway is one of several tools Martian makes externally available alongside ARES (an RL-first framework for training and evaluating LLM agents) and K-Steering (inference-time activation steering).

Update: llm-adapters v8.11.0

The open-source llm-adapters library reached v8.11.0 in May 2025, with the repository last pushed to in July 2025. The project follows semantic versioning and is actively maintained. The hosted Martian Gateway service is the commercial complement to this library, providing the managed API endpoint, dashboard, and broader model catalog.