Cost Waterfall
Tasks cascade through tiers — each request uses the cheapest provider capable of handling it.
Task Router
Describe a task and see which provider tier handles it.
Enter a task above to see routing decision
Provider Stack
Every provider in the orchestration stack and its role.
Primary interactive assistant via OAuth subscription. Used for complex reasoning, architecture, code review, and multi-step tasks.
Large context window for document analysis, long-form generation, and multimodal tasks. Zero marginal cost via subscription.
qwen2.5:7b for trivial tasks, qwen2.5-coder:14b for code generation. Runs on RTX 4080 (16GB VRAM) — zero cost, zero latency.
Llama 3.1 405B and GPT-4o via GitHub Models API. 20,000 requests/day free — ideal for complex tasks when local is insufficient.
Paid tier with high availability and large remaining balance. Routes medium-complexity tasks when GitHub Models is rate-limited.
Real-time library documentation lookup. Retrieves accurate, version-specific API docs and code examples for any framework.
Web-grounded research and real-time information retrieval. Sonar Pro for deep research reports, Sonar for quick lookups.
QLoRA fine-tuning of local models on RTX 4080. HuggingFace Inference API for specialized tasks. Model hosting and versioning.
// Cost Comparison
Real cost per 1M tokens across the provider waterfall.
| PROVIDER | INPUT $/1M | OUTPUT $/1M | TIER | USE CASE |
|---|---|---|---|---|
| Ollama Local | $0.00 | $0.00 | Free | Dev, local reasoning |
| GitHub Models | $0.00 | $0.00 | Free | Inference, fast tasks |
| Gemini Flash | $0.075 | $0.30 | Low | Speed-critical tasks |
| GPT-4o mini | $0.15 | $0.60 | Mid | Balanced quality/cost |
| Claude Sonnet | $3.00 | $15.00 | Premium | Complex reasoning |
| Claude Opus | $15.00 | $75.00 | Elite | Architecture decisions |
Cost Architecture
The waterfall approach routes ~80% of requests to zero-cost providers.