A coding agent for low-cost models that runs commands inside native sandboxing, switches providers/models and harnesses from the terminal, and can test web and native apps through a built-in QA skill.
A coding agent focused on emulating the agent harness that gets the best performance out of low-cost models. It runs commands inside native sandboxing on macOS, Linux, and Windows, switches providers/models and harnesses from the TUI, tests web and native apps through a built-in QA skill, and can run as an Agent Client Protocol agent for editors.
Despite no autonomy signals in the entrypoint, the code-derived surface analysis shows this agent executes arbitrary code and shell commands, drives browsers and native apps via QA skills, and writes to the filesystem. The README describes it as a "coding agent" running commands inside native sandboxing, operating and testing web/native apps autonomously. This is an agentic execution loop that takes multi-step actions (running commands, editing files, driving UIs) without per-step confirmation by design. The presence of a permissions/approvals/sandbox system suggests optional gating, but the default architecture is autonomous action — placing it at Tier 4. It does not modify its own code or spawn agents, so it is not Tier 5.