AI agent evaluation is evolving. Here's what we're building.

Short summary

Humanbound is an open-source testing engine for AI agents that evaluates real behavior—tool calls, multi-turn conversations, API interactions—rather than isolated prompts. Failing tests can be converted into deployable guardrail rules, closing the loop between evaluation and enforcement. It supports fully local execution via Ollama and is seeking community feedback.

• •Open-source agent evaluation engine testing real behavior beyond isolated prompts
• •Failing tests become deployable guardrail rules
• •Supports local air-gapped execution with Ollama

Generated with AI, which can make mistakes.