Documentation
Project overview
ConfigIQ is a Next.js web application for LLM inference sizing, GPU comparison, and cost modeling. The frontend uses same-origin Next.js API routes to reach the AISimulators and aicostings services without exposing service URLs to the browser.
Audience
- ML engineers sizing GPU clusters for inference
- Platform engineers evaluating cloud vs on-premise costs
- Finance and procurement teams modeling GPU spend
Tools
| Tool | Description |
|---|---|
| Performance | Instant GPU memory and throughput estimate |
| Recommend Sizing | Full inference sizing with batching and quantization |
| KV Cache Calculator | Memory breakdown and KV cache capacity analysis |
| GPU Explorer | Side-by-side GPU comparison table and charts |
| Hybrid Savings | Cloud vs on-premise cost modeling |
| Routing Economics | Multi-tier model routing cost analysis |
| Cluster cost | Estimate costs for multi-node GPU clusters |
Guides
- In-app help: search, ask, and take a tour without leaving the page
- User scenarios and sizing paths: which tool to use for which ask, the scenarios we test sizing against, and how to size multiple workloads
- Settings and simulators: simulator, backend, and version choices, and how a shared link behaves
- Costs: hosted, rented, and purchased assumptions
- Model and GPU caveats: known limits and estimates
- Tested-model validation: measured pairs and classifiers
- Pareto validation: predictions checked against measured data
- MCP server: connect Claude and other MCP clients to ConfigIQ
- REST API: endpoint reference and examples
- Contributing: run the app and prepare a change
- DeepWiki: AI-generated, third-party docs from the codebase; not curated documentation