HungRelay · kimi.paytool.spaceAnthropic & OpenAI

Kimi Code relay for teams

Pool sk-kimi keys across api.kimi.com and api.kimi.ai with 3-slot concurrency caps, 60s cooldowns, and hourly client tokens.

Console view for upstream keys and client tokens

Inspect live requests, token usage, and upstream probe checks across your Kimi Code pool before issuing keys to buyers.

Inspect live logs

Pool controls for Kimi Code limits. Key rotation, protocol translation, and hourly tokens.

Caps concurrency per upstream key

Limit each sk-kimi key to 3 concurrent streams and 40 RPM so burst traffic spreads across the pool.

Kimi pool guards
3 of 3 on
  • Cap 3 concurrent streamsPrevents single-account spikes
  • Cooldown 429 for 60sRetries next key in same request
  • Failover .com to .aiSwitches domain on 5xx errors

Translates OpenAI to Anthropic

Accepts OpenAI chat completions and Anthropic messages, including tool calls, thinking blocks, and SSE streams.

JSrelay.mjs SHcurl 2 routes
const host = "api.kimi.com";
const model = "k3.1-pro";
const r = await relay(host, req);
return meter(r, key);
Maps k3.1-pro and kimi-for-coding.

Meters pool tokens and cache

Tracks input, output, and prompt-cache tokens per client key in SQLite WAL with zero external database overhead.

Pool completion rate Last 14 days
99% +5 pts with .com / .ai failover

Issues hourly and quota keys

Create sk-hung keys with 1h, 24h, 7d, or 30d timers and share the self-check portal with buyers.

  • YHyinhe-1h-3k extended 1hQuota 500k tokens · 30 RPM active
  • P2pool-2 entered 60s cooldownHTTP 429 caught · routed to vps-pool-1
  • CCclaude-cli-02 streamed 200 OKk3.1-pro · 410ms · 0 client retries

Endpoints for Claude Code and OpenAI clients

Point Claude Code, Cursor, OpenCode, or custom scripts to kimi.paytool.space without changing client tooling.

  • Anthropic endpoint/v1/messages
  • OpenAI endpoint/v1/chat/completions
  • Storage engineSQLite WAL · 40 MB RAM

View setup snippets

Failover rules for every upstream error

When an upstream sk-kimi key hits a rate limit or timeout, HungRelay switches keys or domains inside the same request.

  • Retry attemptsUp to 3 keys
  • Fallback domainapi.kimi.ai/coding
  • Cooldown window60 seconds
  • HTTP 429Places the upstream key in a 60s cooldown and retries the next healthy key immediately.
  • HTTP 401 / 403Disables the expired upstream key and routes traffic to remaining active keys.
  • HTTP 5xx / TimeoutFails over between api.kimi.com/coding and api.kimi.ai/coding automatically.
  • Concurrency CapHolds each upstream key at 3 active streams to prevent Kimi Code account throttling.
  • Stream AbortCancels the upstream fetch when a client disconnects so slots release right away.
  • Hourly ExpiryEnforces exact millisecond expiration for 1h, 24h, and 7d rental keys.

Questions

Operational details for pool operators and API buyers.

Anthropic OpenAI kimi.paytool.space · /v1

Read portal guide

Which Kimi Code models are supported?

Routes k3.1-pro, kimi-for-coding, and k3.1-pro-highspeed with model alias mapping.

Can buyers use Claude Code and OpenAI SDKs?

Both /v1/messages and /v1/chat/completions accept one sk-hung key with SSE streams.

How does the pool protect upstream accounts?

Each sk-kimi key caps 3 concurrent streams with a 60s cooldown on HTTP 429.

How do buyers check quota and expiry?

Buyers open /portal with their sk-hung key to view tokens and setup commands.

How much RAM does the SQLite relay use?

Node.js 22 with built-in SQLite WAL stays under 60 MB RAM alongside Telegram bots.