future-agi/future-agi

An end-to-end observability and eval platform for LLM apps and AI agents — tracing, evals, simulations, guardrails, and a model gateway in one self-hostable package under Apache 2.0. What separates it from the crowded LangSmith-adjacent space: the simulation layer, which lets you replay agent runs with altered conditions rather than just logging what happened. That is the part worth 30 minutes of your Saturday. The dataset management and guardrails integration mean you are not bolting three separate tools together to go from raw traces to a dataset to a regression test. Honest reservation: the repo is early and the docs are thin in places — expect rough edges if you self-host on day one. But the architecture is sound and the price point for self-hosters is zero, which reframes the LangSmith vs. Braintrust debate entirely. -> Best for: SaaS team of 2-5 shipping LLM-backed features who want observability without a per-seat bill