Experiential Gives AI Agents One Open-Source Gateway to Rule All Models

Experiential is an open source gateway that gives your agents a single OpenAI-compatible endpoint regardless of which model provider sits behind it. The project bundles a router, access control, spend caps, and a traffic-driven fine-tuning loop into one service you can self-host or run as a hosted product. The repo is Apache 2.0, the gateway is free, and the company earns on hosted inference and a Pro tier.

The core premise: one API key covers hosted providers, bring-your-own keys, and local GPUs, with no markup on routed tokens.

Two commands to a running instance

A local instance starts with a pip install and a single run command. A setup wizard asks for a provider, model, and public alias, then prints a one-time virtual key:

pip install experiential
exp run

Any OpenAI SDK works by pointing at http://127.0.0.1:8000/v1 and using the alias, say support-agent, as the model name. The gateway resolves that alias to whichever real model, provider, and fallback chain you configured. Providers wired in out of the box include OpenAI, Anthropic, Gemini, Meta, Qwen, GLM, Kimi, DeepSeek, and Mistral, running on backends like Bedrock, Azure AI Foundry, Fireworks, Modal, Vertex, and OpenRouter.