ActiveHarness
A Ruby framework, made for Rails but usable in plain Ruby, for building observable, cost-controlled LLM pipelines with automatic model fallback, retries and consensus-based decisions.
A Request in ActiveHarness (previously named Agent) wraps a single resilient LLM call with a model fallback chain, retries and hooks; multi-step pipelines can be built from these requests with a Rails-native DSL, grouped into reusable sub-pipelines, and nested, with each sub-pipeline carrying its own stop conditions, context forwarding and execution-time tracking. Deterministic and AI steps can be composed together in one pipeline.
Every request, pipeline step and Tribunal tracks time, tokens and dollar cost. A Tribunal runs several requests in parallel and produces a Verdict based on their agreement, aiming to improve reliability and reduce bias and hallucination through consensus rather than a single model call. It is described as a work in progress, with an API that may still change between versions, and is licensed MIT.