# RelayRank

> RelayRank 是 AI 中转站导航与评分榜单，也提供 Claude、OpenAI/Codex 和 Gemini 兼容端点检测报告。

## Key URLs

- Home: https://ai-relay-rank.vercel.app/
- Claude relay check: https://ai-relay-rank.vercel.app/claude
- OpenAI/Codex relay check: https://ai-relay-rank.vercel.app/openai
- Gemini relay check: https://ai-relay-rank.vercel.app/gemini
- Relay ranking: https://ai-relay-rank.vercel.app/leaderboard
- FAQ: https://ai-relay-rank.vercel.app/faq
- Privacy: https://ai-relay-rank.vercel.app/privacy
- Terms: https://ai-relay-rank.vercel.app/terms
- Contact and takedown: https://ai-relay-rank.vercel.app/contact
- Full LLM context: https://ai-relay-rank.vercel.app/llms-full.txt

## What RelayRank Checks

- Claude relays: official Claude Agent SDK compatibility, fixed marker verification, model/usage evidence, TTFT, latency, and cost metadata.
- OpenAI/Codex relays: Chat Completions and Responses-compatible behavior, tool/function calling, structured output, usage accounting, streaming integrity, model fingerprints, and conversation memory.
- Gemini relays: OpenAI-compatible chat endpoint behavior, model information, tool calling, structured output, usage fields, and response integrity.

## Result Semantics

- Scores summarize evidence from a single detection run; they are not official model certification.
- 85+ generally means key protocol and capability checks passed strongly; 70-84 is passed but still needs risk-note review; 50-69 is marginal; below 50 is failed.
- The public leaderboard aggregates only submitted public detections by domain and protocol; sample count is not traffic volume or SLA evidence.

## API Notes

- Plaintext API keys are not persisted in reports, leaderboard summaries, Blob snapshots, or logs. Durable workers use short-lived encrypted task credentials.
- Reports are public by URL and available as HTML, JSON, and OpenGraph image resources.
- Production deployments should use Postgres storage and distributed rate limiting; the current Vercel-only profile executes jobs in the submitting Node function.