Open-source AI gateway Experiential unifies cloud and local models behind one OpenAI-compatible API and caps agent spend

On September 21, GIGAZINE looked at Experiential, an open-source AI gateway and router published by Experiential Labs that gathers cloud and local models behind a single OpenAI-compatible API and lets operators manage and optimise them for quality, speed and cost.

The problem it targets is familiar. Teams that use GPT, Claude, Gemini and others usually configure a separate endpoint and API key for every provider, which scatters usage tracking and billing across dashboards. Experiential sits in between: an agent configures one OpenAI-compatible API and reaches every provider behind it.

More than 1,000 models ship in the standard catalogue, covering GPT, Claude and Gemini along with DeepSeek, Qwen, Kimi, GLM, Mistral and llama. Users who already hold provider keys can register them with Experiential and keep using them under a bring-your-own-key arrangement.

For companies, the gateway issues a dedicated Experiential API key per employee or per AI agent while the organisation keeps custody of the upstream provider keys. The Logs view records which key made each request in its request history, and filtering shows which models a given key used, so usage and spend can be attributed in one place.

Experiential also optimises model selection from real traffic. Once an agent's activity is accumulated in OpenTelemetry format, the system evaluates a task across several models and compares quality, latency and price to build a router, sending simple work to cheap fast models and hard work to stronger ones.

There are tier limits to note. The cloud Free plan excludes automatic routing and per-prompt model optimisation, which start at the Pro plan, while Free covers the basics of reaching several models through one API and reviewing usage and cost together. Self-hosted deployments can export history accumulated in tools such as Langfuse, in OpenTelemetry or other formats, and load it into Experiential to build a router.

The point of gateways like this is to turn multi-model use from a pile of per-vendor keys into infrastructure that can be observed, capped and optimised. Once agents call tools and spend money on their own, budget control and cost attribution stop being a finance question and become an engineering one.

The open-source gateway lane is getting crowded. GIGAZINE's related coverage on the same page includes Wayfinder, which switches between local and cloud models by task difficulty. Whether Experiential holds up on routing accuracy, self-hosting effort and billing fidelity is the thing to watch.