Experiential — шлюз и роутер для LLM API

★ 8.3K

Experiential is an open-source gateway and router for agent workflows that unifies hosted, BYOK, and local models under a single OpenAI-compatible API and learns to optimize routing from real production traffic. Reach for it when you're tired of paying markups to LLM middleware, when you need one endpoint across multiple providers — OpenAI, Anthropic, Gemini, Azure, Bedrock, Fireworks, OpenRouter — without changing client code, when you want fine-grained control over which agents or users can call which models and how much they can spend, or when you want to turn accumulated traces into a custom router or fine-tuned model. Written in Python, installed via pip install experiential, launched locally with exp, which starts a native data plane on loopback; a managed hosted platform is also available at platform.experientiallabs.ai serving the same API. Traffic capture from Claude Code and Codex desktop apps is supported via exp capture, but it is experimental, requires macOS, and Python 3.13 or newer. Router simulation is built from OpenTelemetry traces; fine-tuning runs through Tinker. The tool routes outgoing LLM calls and optimizes them — it does not host models or serve inbound inference. Best suited to teams with meaningful agent traffic volumes; capture functionality is still under active development.