// maiki.gateway — status: online

ROOT ACCESS.

One gateway to every intelligence

One API key for Claude, GPT, Gemini and 40+ providers. Point Claude Code, Codex or any SDK at Maiki and get one balance, per-key spend caps and automatic failover when a provider goes dark.

Get your key →Read the docs

> claude code // codex // cursor // any sdk

route.loglive

12:04:31claude-sonnet-5-5200 1.2s $0.0042
12:04:33gpt-5-codex200 0.8s $0.0019
12:04:34claude-opus-5-5529 → failover
12:04:34claude-opus-5-5200 2.1s $0.0187
12:04:36gemini-2.5-pro200 1.5s $0.0031

40+

providers

99.9%

uptime

api key

claude-opus-5-5✦claude-sonnet-5-5✦claude-haiku-5-5✦gpt-5✦gpt-5-codex✦o3✦gemini-2.5-pro✦gemini-2.5-flash✦deepseek-v3✦qwen3-coder✦kimi-k2✦glm-4.6✦grok-4✦mistral-large✦

// 01 — plug in your agent

Your agent. Our socket.

Claude Code and Codex speak different protocols. Maiki speaks both natively, so you keep every feature and lose every extra subscription. Two env vars and you're in.

Anthropic-native endpoint. Tool use, thinking and prompt caching pass straight through.

01/04~/.zshrc

1# Point Claude Code at Maiki2export ANTHROPIC_BASE_URL="https://api.maiki.ai"3export ANTHROPIC_AUTH_TOKEN="sk-maiki-••••••••"4 5# then just run it6claude

Anthropic-native endpoint. Tool use, thinking and prompt caching pass straight through.

scroll ↓ to step through

// 02 — system specs

Built for agents that never sleep.

Coding agents fire hundreds of requests per session. Maiki is tuned for that workload, not for a chatbot demo.

Native protocols

Anthropic Messages, OpenAI Chat and Responses, Gemini. No lossy translation layer, so tool calls, thinking and streaming behave exactly like upstream.

Prompt caching intact

Cache headers pass through untouched. Long agent sessions stay cheap, the way they're supposed to.

Auto failover

Upstream throws a 529 at 2am? Maiki retries on a healthy channel before your agent even notices.

One balance

Top up once, burn it on any model. No five separate invoices, no five dashboards.

Per-key limits

Spin up a key per project, teammate or CI job. Cap its spend, restrict its models, revoke it in one click.

Every token logged

Per-request logs with model, latency, tokens and cost. Know exactly what that refactor cost you.

// 03 — boot sequence

Online in 60 seconds.

01
Sign up
Create an account and top up any amount. No subscription, no minimum.
02
Mint a key
Generate an API key and set its spend cap and allowed models.
03
Jack in
Swap the base URL in Claude Code, Codex or your SDK. Ship.

// end of line

STOP JUGGLING KEYS.

One key. Every model. Pay only for the tokens you burn.

Jack in now →

ROOT ACCESS.RROOOOTT AACCCCEESSSS..