Architect
Plan and review
Select a local Ollama model for planning and reviewer duties. The Architect turns the user request into implementation direction and later evaluates the Coder result against both the request and controller evidence.
The idea
Two models. Distinct jobs. One controlled workflow.
OpusAI is built around separation of responsibilities rather than one model improvising every phase of software development. The human user directs the task, an Architect plans and reviews, a Coder implements and repairs, and controller-side validation checks whether the result actually satisfies executable requirements.
USER→ARCHITECT PLAN→CODER IMPLEMENT→CONTROLLER TEST→REVIEW / FIX→FINAL OUTPUT
Coder
Implement and repair
A separate local model handles implementation. OpusAI gives the Coder its own thread and output-token controls so planning and code generation can be tuned independently for the machine and models being used.
Controller
Validate before promotion
Completion checks, Python compilation, and PySide6 offscreen runtime smoke tests provide evidence outside the language models themselves. Failed validation can trigger a correction pass before final review.
Verified delivery
Final Output is earned, not assumed.
The collaboration transcript is not the same thing as a finished result. OpusAI reserves a dedicated Final Output panel for work that survives the controller and review sequence. Successful output can then be copied or saved with Save As.
Local first
Built around Ollama.
Architect and Coder models are selected from a local Ollama environment. OpusAI exposes role-specific CPU thread budgets, role-specific output-token budgets, shared context size, shared temperature, and Ollama keep-alive settings.
Integrated suite
PerformanceLab turns tuning into an experiment.
PerformanceLab is integrated into the main OpusAI process and measures actual local inference behavior instead of guessing from hardware specifications. It records benchmark configuration, cold and warm behavior, repeated measured runs, system state, history and reports. OpusAI can import selected PerformanceLab profile values into runtime settings.