An agent uses a top model for planning and local models for most work
This agent avoids sending every task to an expensive top-tier model. The top model is used mainly for design and planning, while do most of the token-heavy work on a personal dual RTX 3090 machine. The goal is to combine strong reasoning with lower running cost.
such as Qwen 3.5/3.6 27B and are good, but they did not match the judgment and polish of a . The main idea is that software projects often fail or need rewrites because the design is weak, not because typing code is hard. The system uses three swappable tiers controlled by a .
The planner can be Codex or any model that can produce decision JSON, and the setup reuses existing tools instead of rebuilding from scratch.
Key points
- A is used for planning instead of every step of the task.
- handle most of the token-heavy work.
- The setup runs on a personal dual RTX 3090 machine.
- The system has three swappable tiers controlled by a .
- Codex can act as the planner if it outputs decision JSON.