Agent harness
An agent harness is the software wrapped around an AI model that turns it into an agent. The model only predicts text. The harness gives it tools to call, files and memory to work from, rules about what it may touch, and the loop that keeps it working until a job is done. Claude Code, Codex, Cursor and OpenCode are all harnesses.
Why it matters
Two teams can use the same model and get very different results, because most of what makes an agent useful lives in the harness. It decides what context the model sees, which actions it can take without asking, how it recovers from a failed step and when it stops. When people say one coding agent is “better” than another, they are often comparing harnesses, not models.
The harness is also where control sits. Permissions, audit logs and limits on spending are set there, which is why companies care about it as much as engineers do.
A harness is not the same as a model
You can usually swap the model inside a harness. Many harnesses let you pick from several, and some route each task to a different one. That makes the harness the longer-lived choice: models change every few months, while the way your team works with agents (skills, memory files like AGENTS.md, permissions) stays in the harness.
How elvex relates
elvex Enterprise is a company-wide harness that runs on open-weight models, with the controls IT needs. elvex itself doesn’t replace the harness people already use: it connects them, so your Claude Code and a teammate’s Codex can work in the same thread.