Claude, GPT and Grok API Setup Guide

1. What did you receive?

Your order arrives by email and on the My account → Orders page. The delivery is one of three formats:

The addresses and model names in your delivery email always come first: the systems are updated from time to time. If a value on this page differs from your email, use the one in the email and let us know.

2. A key starting with sk- (Claude API · 1-day and 3-day packages)

Works Claude Code (terminal), the Claude Code extension (VS Code, Cursor, JetBrains, Antigravity), Chatbox (provider: Claude), the Anthropic SDK, cURL.

Does not work Codex, Cursor's “Override OpenAI Base URL” setting and extensions set to “OpenAI Compatible”: this key only accepts the Anthropic format.

Setting up Claude Code

1. Install Node.js 18 or newer (nodejs.org; on Windows winget install OpenJS.NodeJS.LTS).
2. Install Claude Code in a terminal:
   Copy

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

3. Open the settings file: macOS / Linux ~/.claude/settings.json, Windows %USERPROFILE%\.claude\settings.json. Create it if it does not exist; otherwise add the env block without deleting your other settings, and replace sk-YOUR-KEY with your key:
   Copy

   {
     "env": {
       "ANTHROPIC_BASE_URL": "https://api.nghimmo.com",
       "ANTHROPIC_AUTH_TOKEN": "sk-YOUR-KEY",
       "ANTHROPIC_MODEL": "nghi/claude-sonnet-5",
       "ANTHROPIC_DEFAULT_OPUS_MODEL": "nghi/claude-opus-5",
       "ANTHROPIC_DEFAULT_SONNET_MODEL": "nghi/claude-sonnet-5",
       "ANTHROPIC_DEFAULT_HAIKU_MODEL": "nghi/claude-haiku-4.5",
       "ANTHROPIC_SMALL_FAST_MODEL": "nghi/claude-haiku-4.5",
       "CLAUDE_CODE_USE_GATEWAY": "1",
       "CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY": "1",
       "CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK": "1",
       "API_TIMEOUT_MS": "600000"
     }
   }

   If ANTHROPIC_API_KEY is set, leave it empty; only ANTHROPIC_AUTH_TOKEN is used.
4. Open a new terminal, go to your project folder and type claude. If it asks how to sign in, choose the API key option; do not sign in with an Anthropic account.
5. Switch models inside a session: /model nghi/claude-opus-5. Opus and Fable use tokens faster; Sonnet is enough for everyday work.

VS Code, Cursor, JetBrains, Antigravity: install the Claude Code extension; it reads the same settings.json. Fully quit and reopen the editor after changing the settings (or Reload Window).

If a model with [1m] (1M context) in its name fails, use the same model without [1m].

Chatbox (desktop and web)

1. Settings → Model Provider → Claude.
2. API Host https://api.nghimmo.com/v1, API Key your key.
3. Add the model nghi/claude-sonnet-5 and press Test.

Quick test (cURL) and Python

curl https://api.nghimmo.com/v1/messages \
  -H "x-api-key: sk-YOUR-KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"nghi/claude-haiku-4.5","max_tokens":64,"messages":[{"role":"user","content":"Hello"}]}'

import anthropic

client = anthropic.Anthropic(base_url="https://api.nghimmo.com", api_key="sk-YOUR-KEY")
msg = client.messages.create(
    model="nghi/claude-sonnet-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
print(msg.content[0].text)

3. A code starting with TSL- (GPT API · Codex)

First, turn the code into a key

1. Open redeem.teamsoclo.site, paste the code and get a new key (“Get a new key”; if the page opens in Vietnamese, “Đổi ra key mới”).
2. Copy the sk-… key with the Copy button and keep it safe. Selecting it with the mouse can drop a character at either end, which later shows up as a 401 error.
3. The package validity starts at this moment: redeem the code when you are ready to use it. If you lose the key, enter the same code again and the same key appears.
4. When you buy a new code of the same package, “Top up an existing key” (Vietnamese “Nạp vào key có sẵn”) adds its time and credit to your existing key.

Redeeming works the same for the GPT API, Grok API and 1-month Claude API packages. With a Claude code go to section 4, with a Grok code to section 5; the settings below are for GPT.

Works Codex CLI, the VS Code Codex extension, the Codex app, Cursor (Chat), Cline, Roo Code, Kilo Code, Continue, Chatbox, the OpenAI SDK, cURL. The gateway also translates the Anthropic format, so it can be used in Claude Code too.

Does not work Cursor's Tab completion and Composer Agent: Cursor runs those on its own infrastructure.

Codex CLI and the VS Code Codex extension

1. With Node.js 18+ installed:
   Copy

   npm install -g @openai/codex

2. Write ~/.codex/config.toml (Windows %USERPROFILE%\.codex\config.toml) like this:
   Copy

   model = "gpt-5.6-sol"
   model_provider = "apigw"
   
   [model_providers.apigw]
   name = "API"
   base_url = "https://gpt.teamsoclo.site/v1"
   env_key = "APIGW_KEY"
   wire_api = "responses"
   requires_openai_auth = false

   Keep wire_api = "responses"; changing it to chat breaks Codex.
3. Put the key in the APIGW_KEY environment variable:
   Copy

   # Windows (PowerShell or CMD)
   setx APIGW_KEY "sk-YOUR-KEY"
   
   # macOS / Linux (~/.bashrc if you use bash)
   echo 'export APIGW_KEY="sk-YOUR-KEY"' >> ~/.zshrc

4. Open a new terminal and type codex. The VS Code extension reads the same settings; quit VS Code completely and reopen it (Reload Window is not enough).

Cursor

1. Ctrl+Shift+J (macOS Cmd+Shift+J) → Models.
2. Turn on Override OpenAI Base URL, enter https://gpt.teamsoclo.site/v1, paste the key and press Verify.
3. Add gpt-5.6-sol with “+ Add model” and turn off all of Cursor's default models. If they stay on, requests go to real OpenAI with your key and you see an “invalid key” error.

Cline, Roo Code, Kilo Code and Chatbox

If you get 404, try the Base URL without /v1: some versions add /v1 themselves and the address becomes /v1/v1.

Claude Code (optional)

The gateway translates the protocol, so the GPT key also works in Claude Code; the model used is still GPT. The env block in settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://gpt.teamsoclo.site",
    "ANTHROPIC_AUTH_TOKEN": "sk-YOUR-KEY",
    "ANTHROPIC_MODEL": "gpt-5.6-sol",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "gpt-5.6-sol",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "gpt-5.6-sol",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "gpt-5.6-sol",
    "ANTHROPIC_SMALL_FAST_MODEL": "gpt-5.6-sol"
  }
}

OpenAI SDK (Python) and reasoning level

from openai import OpenAI

client = OpenAI(base_url="https://gpt.teamsoclo.site/v1", api_key="sk-YOUR-KEY")
r = client.chat.completions.create(
    model="gpt-5.6-sol",
    reasoning_effort="medium",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

Without a reasoning level the gateway uses a high one and the credit runs out fast. For simple work choose medium or low: reasoning_effort in Chat Completions, "reasoning": {"effort": "medium"} in the Responses API.

4. TSL- code: 1-month Claude API packages

The 10M tokens · 1 month and 50M tokens · 1 month packages arrive as a TSL- code. Redeem it at redeem.teamsoclo.site as in section 3; the 1-month validity starts at that moment.

Works Claude Code (terminal and the VS Code / JetBrains extension), Codex, Cursor (Chat), Cline, Roo Code, Kilo Code, Chatbox, the Anthropic and OpenAI SDKs. This key calls only claude-… models.

Claude Code

Install Claude Code as in section 2 (Node.js 18+ and the npm command), then write this into settings.json (keep your other settings):

{
  "autoCompactWindow": 250000,
  "env": {
    "ANTHROPIC_BASE_URL": "https://gpt.teamsoclo.site",
    "ANTHROPIC_AUTH_TOKEN": "sk-YOUR-KEY",
    "ANTHROPIC_MODEL": "claude-opus-5",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-5",
    "ANTHROPIC_SMALL_FAST_MODEL": "claude-haiku-4-5",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

autoCompactWindow: Claude Code treats the newer models as 1M-context models, while the gateway accepts about 280K tokens per turn. Without 250000 long sessions stop with “Autocompact is thrashing”; typing /autocompact 250k once in an open session does the same. If ANTHROPIC_SMALL_FAST_MODEL is left empty, Claude Code calls a background model that is not in your package.

Cursor, Cline, Chatbox and Codex

Anthropic SDK (Python)

import anthropic

client = anthropic.Anthropic(base_url="https://gpt.teamsoclo.site", auth_token="sk-YOUR-KEY")
msg = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
print(msg.content[0].text)

5. TSL- code: Grok API

The Grok API packages (3 days, 5 days or 1 month) arrive as a TSL- code. Redeem it at redeem.teamsoclo.site as in section 3; the package validity starts at that moment.

Works Cursor (Chat), Cline, Roo Code, Kilo Code, Continue, Chatbox, the OpenAI SDK, cURL; Claude Code and Codex too, because the gateway translates the protocol. This key calls only grok-… models; any other model returns “No available channel”.

Does not work Cursor's Tab completion and Composer Agent; the xAI console and xAI's own apps (this is not an official xAI key).

Cursor

1. Ctrl+Shift+J (macOS Cmd+Shift+J) → Models → turn on Override OpenAI Base URL, enter https://gpt.teamsoclo.site/v1, paste the key and press Verify.
2. Add grok-4.7 with “+ Add model” and turn off all of Cursor's default models.

Cline, Roo Code, Kilo Code and Chatbox

Claude Code and Codex

The env block in Claude Code's settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://gpt.teamsoclo.site",
    "ANTHROPIC_AUTH_TOKEN": "sk-YOUR-KEY",
    "ANTHROPIC_MODEL": "grok-4.7",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "grok-4.7",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "grok-4.7",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "grok-4.7",
    "ANTHROPIC_SMALL_FAST_MODEL": "grok-4.7"
  }
}

In Codex use the config.toml from section 3 and only set model = "grok-4.7".

Quick test (cURL) and Python

curl https://gpt.teamsoclo.site/v1/chat/completions \
  -H "Authorization: Bearer sk-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"grok-4.7","messages":[{"role":"user","content":"Hello"}]}'

from openai import OpenAI

client = OpenAI(base_url="https://gpt.teamsoclo.site/v1", api_key="sk-YOUR-KEY")
r = client.chat.completions.create(
    model="grok-4.7",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

6. A code starting with BDK- (Claude API · 15M and 40M tokens, 1 day)

The 15M tokens · 1 day and 40M tokens · 1 day packages arrive as a BDK- code. The validity starts when you redeem the code and lasts 24 hours. Tokens are used at a different rate per model: Opus and Fable faster, Sonnet and Haiku slower.

1. daily.bddevlab.buzz/redeem.html → Redeem tab → paste the code. The key (sk-…), credit amount, validity and expiry time appear; copy the key with Copy.
2. The validity starts when you redeem the code. Entering the same code again shows the same key; do not share the code or the key.
3. Credit left, time left and recent requests are in the Usage tab of the same page.

The env block in settings.json for Claude Code (adjust the model names to your list):

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.bddevlab.buzz",
    "ANTHROPIC_AUTH_TOKEN": "sk-YOUR-KEY",
    "ANTHROPIC_MODEL": "claude-sonnet-5",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4.5",
    "ANTHROPIC_SMALL_FAST_MODEL": "claude-haiku-4.5"
  }
}

Cline, Roo, Kilo and Chatbox: provider OpenAI Compatible, Base URL https://api.bddevlab.buzz/v1, model name from the list.

If a model does not answer (for example the newest Opus version when it is busy), pick another model from the list; your remaining credit is in the Usage tab.

7. Checking credit and time left

When the time or the credit runs out, requests are refused with 401 or “insufficient quota”; unused credit does not move to a new package.

8. Common errors and fixes

The quickest way to isolate a problem: validate the key on the check page, then run the cURL test above. If cURL works but the app does not, the problem is in the app settings.

9. Saving tokens

• Start a new chat when the topic changes: every message resends the whole history, so message 50 of a long chat costs many times more than the first.
• In Claude Code use /clear (new task) and /compact (every 30–50 messages).
• Point the agent at the files it needs (@file) instead of letting it scan the whole project; turn off Codebase Indexing in Cursor.
• Use a lighter model for simple work (Haiku, Sonnet); Opus and Fable use more credit.
• For GPT, lower the reasoning level (medium / low).

10. Security and usage rules

• Do not share the key or the code. The per-minute limit and the credit are shared by everyone who uses the key.
• When asking for support, write only the last 4 characters of the key.
• The key is not an official Anthropic, OpenAI or xAI console key; your requests pass through an API gateway. Do not send passwords, personal data or secret source code.
• If the instructions in your delivery differ from this page, follow the delivery.

11. Warranty and support

Warranty: if the key does not work within the package period, you get a new key or a refund for the unused part. Details are on the product page.

Support: AIStore Assistant, on the left edge of the site, answers setup questions right away. If it is not solved, email info@aistoreplus.com with your order number, a screenshot of the error and the last 4 characters of the key.