Choose a Starting Point
Pick a preset that matches your goal — each pre-loads a coherent architecture. Personal Superintelligence balances power and breadth; Local Privacy First runs entirely on-device; Developer Copilot optimizes for code; Creative Partner maximizes reasoning and context.
Customize Components
Add, remove, or reorder components in the left panel. Each component type — Model, Memory, Tools, Runtime — has configurable parameters. Changes reflect instantly in the diagram, radar, and metrics.
Read the Radar
The radar chart scores six dimensions 0–100: Reasoning, Coding, Creativity, Knowledge, Speed, Privacy. No single stack maximizes all — the shape reveals your tradeoffs.
Inspect Latency
The waterfall breaks down end-to-end latency by stage: input processing, retrieval, model inference, tool calls, output formatting. Hover a bar to see the component responsible.
Export & Share
Download a .json specification for programmatic use, or .md for documentation. The share link encodes your entire stack in the URL — paste it to recreate the architecture.
Limitations & Assumptions
- Latency estimates assume typical 2024 hardware; actuals vary by provider, region, and load.
- Cost uses public list prices where available; enterprise discounts not modeled.
- Capability scores are heuristic composites, not benchmarks.
- Local models require sufficient VRAM/RAM; the tool warns but cannot enforce.