Aibeat

AI Beat is an open source suite for testing AI behavior before it reaches users, published by tophant.ai. It is organized around separate products for separate system boundaries: PromptBeat red teams and evaluates models and prompts, AgentBeat evaluates black box HTTP agents, and RouterBeat publishes live security benchmarks of AI relay platforms. It is built for developers, security teams and evaluators who need evidence rather than impressions. PromptBeat is a Go command line tool that uses Promptfoo as its current evaluation backend; it generates and inspects test cases locally without uploading sensitive data, covers prompt injection, jailbreak resistance and guardrail regressions, and exports standalone HTML reports with failure cases and scoring details. AgentBeat runs on its own native Go runtime, calls an agent over HTTP without changes to its internal code, and scores goal completion alongside optional runtime evidence, tool traces and independently verified state changes, combining deterministic checks with an LLM judge. RouterBeat grades relay platforms into S, A, B and C tiers across stability, response integrity, model fingerprints, billing transparency and prompt leakage. An AIBeat Skill loads into a coding assistant so users can describe a test in plain language and have the assistant plan it and generate offline cases for PromptBeat. No pricing is published. PromptBeat version 0.4.1 is a free download for Linux, macOS on Apple Silicon and Intel, and Windows, and AgentBeat is available as a website evaluation preview with a downloadable build. AgentBeat is still marked as a preview, with its official GitHub evaluator release pending, so only PromptBeat is a fully public package today. The vendor also cautions that some models refuse security test inputs and that testing must stay within authorized scope.