Experiential Labs

Teams building with AI need to use many different model providers (OpenAI, Anthropic, Gemini, local GPUs, self-hosted models) but juggling separate keys, endpoints, and pricing for each is costly and hard to track. Without a unified layer, organizations lack visibility into spend by agent, person, or model, and have no easy way to enforce budget caps or know when switching models or improving caching would save money. Companies also struggle to safely roll out fine-tuned or self-hosted models alongside hosted providers behind a consistent interface.

Experiential Labs offers an open source AI gateway, written in Rust, that lets users point agents at a single endpoint (api.experientiallabs.ai/v1) and route to any model, including hosted providers, a user's own API keys, or their own GPUs, at the provider's price with zero markup on routed tokens. An intelligence layer sits on top of the gateway, watching traffic to recommend model switches, surfacing caching opportunities (with repeated tokens served at 90% off when enabled), and supporting fine-tuning a model on a customer's own traffic that is proven in simulation before serving live. A console gives attribution and spend visibility by agent, person, model, or day, plus request logs showing routes, fallbacks, and provider pricing. Admin-set API keys support caps by day, week, or month, role scopes, model allowlists, and local-only scopes, and the gateway can be self-hosted or used as a hosted service.