Lab feature · no request proxying

AI Token Router Config Lab

Turns TokenAPI Scan pricing, model coverage and availability evidence into LiteLLM / OpenRouter config drafts. This page only generates config drafts — it does not accept or store user API keys, and does not forward real API requests.

How ranking works: for a given model, TokenAPI Scan collects each provider channel's live price per 1M tokens, pass-rate from periodic integrity probes, protocol compatibility and endpoint health, then scores each candidate on cost / integrity / availability / coverage / risk and recommends a role — primary, backup, watch, or avoid. The same intelligence is available as a machine-readable JSON API: GET /api/router/candidates?model={model_id}&limit=N (returns generated_at, CC-BY-4.0 licensed; cite “TokenAPI Scan” when referencing).

Currently a lab capability: useful for validating data scope and config shape first; not yet promoted to the homepage main entry.

Model canonical id Target router Candidate count

Sample matrix: one-click validation of gpt-4o / claude-sonnet-4 / gemini-2.5-pro / deepseek-r1 / qwen3-32b × 4 targets — only checks candidate resolution status, no real requests sent.

Click "Generate config draft" to see YAML / JSON here.