Quickstart

One command. Interactive wizard. You're live in 60 seconds.

Run the install script. It detects your shell, installs the CLI, writes the config, and verifies your key — all in one go.

macOS · Linux

curl -fsSL https://app.alphayellowish.com/setup.sh | bash

Windows (PowerShell)

irm https://app.alphayellowish.com/setup.ps1 | iex

Try it

codex "write a python function that reverses a string"

Prefer manual setup?

If you'd rather not pipe a script, the Codex CLI setup page has the manual config you can copy-paste.

Codex CLI

OpenAI's official terminal client. The recommended way to use Alphayellowish.

If you followed the Quickstart, you're already done. If you want to install manually:

npm install -g @openai/codex

Claude Code CLI

Anthropic's terminal tool, pointed at Alphayellowish for GPT-5.5 access.

If you prefer Claude Code's interface but want OpenAI's GPT-5.5 as the model, you can point Claude Code CLI at our endpoint.

npm install -g @anthropic-ai/claude-code

export ANTHROPIC_BASE_URL="https://app.alphayellowish.com"
export ANTHROPIC_API_KEY="sk-your-key-here"
export ANTHROPIC_MODEL="gpt-5.5"

claude

Add the three export lines to your ~/.zshrc to make them permanent.

VS Code

Two ways: the Codex extension or direct settings.json.

Option A — Codex extension

Option B — settings.json

{
  "codex.baseUrl": "https://app.alphayellowish.com/v1",
  "codex.apiKey": "sk-your-key-here",
  "codex.model": "gpt-5.5"
}

Cursor

Override Cursor's OpenAI base URL in its model settings.

Windsurf

Configure the OpenAI Compatible Provider in advanced settings.

Cline / Roo Code

Both extensions share the same configuration pattern.

Continue.dev

Edit your ~/.continue/config.yaml to add Alphayellowish as a model.

models:
  - name: GPT-5.5 (Alphayellowish)
    provider: openai
    model: gpt-5.5
    apiBase: https://app.alphayellowish.com/v1
    apiKey: sk-your-key-here

Aider

Aider works with any OpenAI-compatible endpoint via environment variables.

export OPENAI_API_BASE="https://app.alphayellowish.com/v1"
export OPENAI_API_KEY="sk-your-key-here"

aider --model gpt-5.5

Add the two export lines to ~/.zshrc or ~/.bashrc for persistence.

Any OpenAI-compatible client

If your tool supports a custom OpenAI base URL, it works with us. Use these three values:

Tested working with

LibreChat, OpenWebUI, ChatBox, NextChat, BetterChatGPT, Lobe Chat, BoltAI, TypingMind, Simon Willison's llm CLI, n8n OpenAI nodes, and Zapier OpenAI integrations.

If your tool needs anything beyond these three values, reply to your welcome email and we'll help.

Reading your usage

Visit app.alphayellowish.com/dashboard, paste your key, bookmark the page.

The dashboard shows what your plan allows, what you've used, and what's left — refreshed every 30 seconds.

What you'll see

• Tokens used in the current 5-hour window — your burst-rate consumption
• Time until reset — when the oldest tokens roll off and free up cap
• Today's request count — how active you've been
• Hourly breakdown — usage curves over the last 24 hours
• Cached vs fresh tokens — how much you've saved via prompt caching

The 5-hour rolling window

Every plan caps usage in 5-hour rolling windows — the same pattern used by OpenAI Codex Pro and Anthropic Claude Max.

The window slides continuously. If you used 200K tokens 3 hours ago and 100K tokens 30 minutes ago, both count toward your current ceiling. Older usage rolls off automatically as time passes.

Why a rolling window?

Real engineering work is bursty — you might code intensely for 2 hours, then idle for 6. A daily cap punishes the burst-then-rest pattern. A 5-hour rolling window rewards it: heavy use followed by a break gives you cap back faster than a fixed-day reset.

Hitting the cap

If you exceed the 5h window, the next request returns HTTP 429 with a clear "resets in X minutes" message. The Retry-After header tells you exactly when to retry. No surprise overage charges.

Authentication

OpenAI-compatible REST API. Drop-in for any OpenAI SDK.

POST /v1/chat/completions

The primary endpoint. Returns a model response for a given chat conversation.

Request

curl https://app.alphayellowish.com/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [
      {"role": "user", "content": "Write hello world in Rust"}
    ],
    "max_tokens": 200
  }'

Response

{
  "id": "resp_0b234…",
  "object": "chat.completion",
  "model": "gpt-5.5",
  "choices": [{
    "index": 0,
    "finish_reason": "stop",
    "message": {
      "role": "assistant",
      "content": "fn main() {\n    println!(\"Hello, world!\");\n}"
    }
  }],
  "usage": {
    "prompt_tokens": 47,
    "completion_tokens": 22,
    "total_tokens": 69
  }
}

Streaming responses

Add "stream": true to receive Server-Sent Events.

The smart proxy correctly tees streaming chunks AND records final token usage on stream end — your quota counter stays accurate.

from openai import OpenAI

client = OpenAI(
    base_url="https://app.alphayellowish.com/v1",
    api_key="sk-your-key",
)

stream = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "user", "content": "Tell me a joke"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Rate limits

Every response includes headers so your client can self-pace.

429 response

Returned when you've exceeded the 5h window cap.

HTTP/1.1 429 Too Many Requests
Retry-After: 14512
X-RateLimit-Limit: 1000000
X-RateLimit-Remaining: 0
Content-Type: application/json

{"error":{"message":"5h token quota exceeded","type":"rate_limit","code":"429"}}

Available models

Two production models, both routed to real OpenAI infrastructure.

Both support streaming, system messages, multi-turn conversations, and JSON mode. We do not silently swap models — what you request is what you get.

401 Unauthorized

Three causes, in order of likelihood:

429 Too Many Requests

You've hit your 5-hour window cap.

The Retry-After response header tells you seconds to wait — divide by 60 for minutes.

The cap resets gradually, not all at once. Your oldest tokens roll off the rolling window first. If you used 900K tokens 4.5 hours ago and 100K tokens 30 minutes ago, the 900K rolls off in ~30 minutes, freeing up most of your cap.

If you're hitting 429 frequently, you may need a higher tier. Reply to your account email and we'll suggest the right plan based on your dashboard's burn rate.

Model not found

You requested a model name not in your plan's allowlist.

The two supported models are gpt-5.5 and gpt-5.3-codex. We do not support Claude, Gemini, or older OpenAI models.

If you're using Codex CLI and seeing this error, check that wire_api = "responses" is set in ~/.codex/config.toml (older Codex CLI versions used "chat_completions" — newer versions only accept "responses").

Quota burning faster than expected

Most often this is prompt-cache misses on long-running sessions.

If your tokens are draining faster than expected, the usual cause is your CLI restarting the conversation cache. Check your dashboard — if "cached vs fresh tokens" shows mostly fresh, you're losing the cache benefit. Restart a fresh session, keep prompts small, and avoid switching models mid-conversation.

If that doesn't help, email us with your dashboard screenshot — we'll diagnose.

Verify your setup

One copy-paste command that tells you exactly what's working or what's broken.

KEY="sk-your-key-here"
CODE=$(curl -sS -o /dev/null -w "%{http_code}" \
  https://app.alphayellowish.com/v1/chat/completions \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.5","messages":[{"role":"user","content":"pong"}],"max_tokens":3}')
case "$CODE" in
  200) echo "Setup works." ;;
  401) echo "Invalid key. Re-copy from your email." ;;
  429) echo "Quota exceeded. Wait for 5h window to reset." ;;
  404) echo "Model not in plan. Use gpt-5.5 or gpt-5.3-codex." ;;
  *)   echo "HTTP $CODE — email us with this output." ;;
esac

Frequently asked questions

Is this real OpenAI or a swapped model?

Real OpenAI. Every request routes to GPT-5.5 or gpt-5.3-codex on OpenAI infrastructure. We never silently substitute models.

Do you log my prompts?

No. Only token counts are recorded for quota enforcement. Prompts and responses stream through and are not persisted.

Can I cancel anytime?

Yes. Monthly billing, no lock-in. Your key stays active till the end of your paid month after cancellation.

What if I exceed my cap mid-task?

The in-flight request completes normally. The NEXT request returns 429 with a clear "retry-in" message. No surprise charges, no truncated responses.

How do I add team members?

Each member gets their own key with their own dashboard. Email us with employee emails plus which tier each.

Invoicing in INR?

Yes. Invoices in INR by an India-registered LLP with GSTIN — suitable for input-tax-credit claims.

What if OpenAI changes model names?

Our identifiers (gpt-5.5, gpt-5.3-codex) stay stable. If OpenAI deprecates a model, we map you to the closest replacement and notify you.

Plans

Two tiers, all with honest caps. Real numbers, verifiable in your dashboard.

Pro · 5×

Real solo-dev workload — 3-5 hours of heavy coding daily, every day of the month.

Team · 20×

Full-time engineer pace — sustained 6-8 hours of heavy use daily. Matches OpenAI Codex Pro 20× and Anthropic Claude Max 20× caps.