Creating highly efficient agents: 450M tool-calling tokens distilled for post-training from top open-source models

Harnesses If you’ve used Claude Code or Codex , you’ve used a harness. A harness is the infrastructure layer that wraps an AI coding agent and decides how it operates, what it can touch, and how you measure whether it worked. It’s how most engineers interact with frontier models today. The category is moving fast. OpenClaw crossed 340,000 GitHub stars in six months, making it one of the fastest-growing repositories on the platform. Hermes Agent is on a similar trajectory. A general-purpose harness can wire your agent into the tools you already use: iMessage, Discord, and others. The interesting part for engineers running their own stack: many of these harnesses can pair with open models that are small enough to run on a single GPU at home and hold up under real workloads.