Is OpusAI a general chat client?
No. Its primary design is a human-directed collaborative development workflow with distinct Architect and Coder roles, controller validation, review/correction passes and a verified Final Output area.
Does OpusAI require cloud AI APIs?
The documented runtime is local Ollama. Architect, Coder and PerformanceLab workflows are built around models available through the configured local Ollama environment.
Do I need two different models?
No. OpusAI provides separate role selectors, which means you can choose different models when that is useful, but the same installed Ollama model can be assigned to both roles.
Why separate Architect and Coder roles?
The separation gives planning/review and implementation different responsibilities and independent runtime budgets. It also makes the review loop more explicit than asking a single response to plan, code and declare itself correct.
What does the controller do?
The controller provides validation evidence outside the language models. Current documented behavior includes completion checks, Python compilation validation and PySide6 offscreen runtime smoke testing. The agent instructions treat controller results as ground truth.
Does a successful compile guarantee the application is correct?
No. Compilation and smoke tests catch important classes of failure, but they do not prove every behavior, UI state, business rule or platform interaction is correct. The human remains the final authority.
What happens when validation fails?
The failed implementation can be sent through a correction pass. The Coder is instructed to repair concrete controller failures before optional refactoring, followed by retesting and review.
Where does finished work appear?
Work that clears the relevant validation/review sequence is promoted to the dedicated Final Output panel, where it can be copied or saved with Save As.
Can I tune Architect and Coder differently?
Yes. The source exposes separate CPU-thread and output-token settings for the two roles while sharing context size, temperature and Ollama keep-alive.
What is PerformanceLab?
PerformanceLab is OpusAI's integrated local LLM benchmark and system-assessment utility. It measures actual Ollama inference on the current machine and preserves configuration, system state, repeated results, history and reports for controlled comparisons.
Does PerformanceLab automatically optimize my computer?
No. Its philosophy is controlled measurement. It can explicitly change the Linux CPU governor when the user requests it, but it is not designed to silently change many settings and call the result optimized.
Why does PerformanceLab use cold, warmup and measured runs?
Cold loading, runtime warmup and sustained generation answer different performance questions. Keeping them separate prevents model-load behavior from being confused with already-resident inference throughput.
Can a small benchmark gain be noise?
Yes. Scheduler activity, background services, cache state, temperature, I/O and memory pressure can all produce ordinary run-to-run variation. Individual runs and environmental conditions should be examined before accepting a small percentage change as an optimization.
Where are benchmark records stored?
config/performance_lab.json data/performance_lab/benchmarks.json reports/
Does PerformanceLab need root?
Normal scanning and benchmarking do not require the whole application to run as root. CPU-governor changes request elevated permission with pkexec only when explicitly selected.
Why is the website download source-only?
The current project distribution on this site is intentionally the OpusAI 1.2 source archive. The project includes a local PyInstaller build script for users who want to compile a one-file application on their own machine.
What operating system is the project focused on?
Linux. PerformanceLab in particular uses Linux system interfaces and utilities for hardware, thermal and CPU-governor inspection.
What happens when I click the window X?
The documented desktop behavior hides OpusAI to the system tray. Use the tray menu to show it again or quit the suite.