OpenAgent Routes One Task Across 173 AI Workers

Field notes

OpenAgent presents routing as a feedback loop rather than a one-time model choice. Its documented sequence analyzes task type and complexity, scores reachable workers, executes in a sandbox, then checks the result against schemas, constraints, tests, and a model judge. The catalog currently names 173 workers across 146 models and three agent harnesses, including Claude Code, Codex, and OpenCode. The example integration uses a bearer token and sends a plain-language programming task to /v1/tasks.