Connect your tool in a minute.

1. Create a key

   In the dashboard, pick a model and click Create key. Copy the key right away: it's shown only once.

2. Set up your tool

   Find your tool below. Right after you create a key, the dashboard shows the same steps with your key already filled in.

3. Start working

   Use your tool as usual. Each request is paid from your balance, and the dashboard shows it with its cost.

Each key works with one model: the one you pick when you create it. That model answers every request made with the key, whatever model name your tool sends. Want two models? Create two keys.

For the same reason, GET /v1/models lists only your key's model. Tools that show a model list, like Cline and Roo Code, offer just that one.

All models and their prices are on the pricing page.

You add money to your balance first. New accounts start with $50 of welcome credit, one per person. Each request costs its tokens at your model's price, and that amount comes off the balance. The dashboard lists every request with its tokens and cost.

When requests stop

When you create a key, you can also choose these two settings. The dialog shows what the input of a prompt that fills the chosen window costs.

The setup steps in the dashboard pass both settings on to your tool.

1. Open the settings

   In Cursor, open Settings → Models and scroll down to API Keys.

2. Paste your key and the base URL

   Paste your key into OpenAI API Key. Turn on Override OpenAI Base URL and paste the base URL.

   OpenAI API Key            sk-tlc-your-key
   Override OpenAI Base URL  on
   Base URL                  https://api.tokenlowcost.com/v1

3. Add a model name

   Click + Add model, type a name like tlc-opus and turn it on. The name is just a label: your key decides the model. Use a name Cursor doesn't already have, so the request goes out with your key.

4. Check and chat

   Click Verify next to the key. Then pick your model in the chat.

The chat built into VS Code can use your key. You need VS Code 1.122 or newer. If VS Code asks you to sign in when it first starts, click Continue without Signing In.

1. Add a custom endpoint

   Press Ctrl + Shift + P (Cmd on a Mac) and run Chat: Manage Language Models. Click Add Models, then Custom Endpoint.

2. Enter a name and your key

   Name it TokenLowCost and paste your key. At API Type, just press Enter.

3. Paste your model

   VS Code opens a file with an empty model in it. Replace its "models": [ … ] part with the block below and save. The dashboard gives you this block ready-made for your key, right after you create it. Copying from here? Take the Claude block for Claude models, the other one for the rest, and change id and name to your model.

4. Pick the model

   Open the chat, click the model name in the message box and pick your model.

Not signed in to GitHub? VS Code then shows Set BYOK utility models. Click Configure and choose Main Agent Model, so chat titles and commit messages use your model too. These are small requests, billed like any other.

Claude Code or Codex in VS Code

The Claude Code and Codex extensions use the same settings as their terminal versions. Paste the Claude Code or Codex line first, then install the extension and restart VS Code. For Claude Code, also open Settings, search for Claude Code login and tick Disable Login Prompt.

One line saves the base URL, your key and your key's model, and every new terminal picks them up. The line below is for a Claude Opus 5.5 key and has a placeholder key. The dashboard shows it made for your key: its model, context window and effort.

1. Paste the line

   Paste it into PowerShell (Windows) or Terminal (macOS, Linux) and press Enter.

2. Run Claude Code

   Run claude in the same window. The first time, press Enter on the text style and security screens. When it asks whether you trust the folder, press ↓ to choose Yes, I trust this folder, then Enter: the preselected No, exit closes Claude Code. You don't need an Anthropic account.

Why the model name? Without it, Claude Code thinks it's talking to Claude Opus 5.5, whatever your key's model is. It tells the model so and signs commits with that name. With the name, it knows the real model. For a model that isn't Claude, Claude Code may say it doesn't recognize the model. That's fine: it only knows Claude models by name, and the requests still work.

Rather use a file? Put this in ~/.claude/settings.json (on Windows, %USERPROFILE%\.claude\settings.json):

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.tokenlowcost.com/v1",
    "ANTHROPIC_AUTH_TOKEN": "sk-tlc-your-key",
    "ANTHROPIC_MODEL": "claude-opus-5-5[1m]"
  }
}

Set ANTHROPIC_MODEL to your key's model: claude-opus-5-5, claude-sonnet-5, claude-haiku-4-5, or for other models the id from the price list, such as gpt-6-sol. Add [1m] at the end if the key's window is over 200K.

Claude Code also sends small background requests, such as chat titles. They use your key's model and are billed like any other request.

One line writes the Codex settings file and saves your key. If you already have a Codex config, it's kept as config.toml.bak. This line has a placeholder key; the dashboard shows it with yours.

Paste it into PowerShell (Windows) or Terminal (macOS, Linux) and press Enter. Then run codex in the same window and choose Trust and continue. You don't need a ChatGPT account.

Rather edit the file yourself? This is what the line writes to ~/.codex/config.toml. If you add it to an existing file, keep the first lines above any [section].

Codex reads your key from the TOKENLOWCOST_API_KEY environment variable, which the line also sets. Set it yourself too: add export TOKENLOWCOST_API_KEY=sk-tlc-your-key to ~/.zshrc or ~/.bashrc, or on Windows run setx TOKENLOWCOST_API_KEY sk-tlc-your-key and open a new window.

In Cline or Roo Code, open the provider settings and fill in the fields below. The first time you open Cline, choose Bring my own API key to get there. Once the key is in, the model list shows just your key's model: pick it.

API Provider  OpenAI Compatible
Base URL      https://api.tokenlowcost.com/v1
API Key       sk-tlc-your-key
Model ID      anthropic/claude-opus-5.5

Cline shows the cost as $0, since it doesn't know our prices. The real cost is under Usage in your dashboard.

For Continue, put this in ~/.continue/config.yaml in place of what's there. Already have models in it? Add just the TokenLowCost entry under models:. The tool_use line lets Agent mode edit files; without it, Continue can't tell the model has tools.

name: My Config
version: 0.0.1
schema: v1
models:
  - name: TokenLowCost
    provider: openai
    model: tlc-opus
    apiBase: https://api.tokenlowcost.com/v1
    apiKey: sk-tlc-your-key
    capabilities:
      - tool_use

Any other tool with an OpenAI Compatible option works the same way.

Set the base URL and your key. The rest is the usual SDK code.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.tokenlowcost.com/v1",
    api_key="sk-tlc-your-key",
)

r = client.chat.completions.create(
    model="claude-opus-5.5",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.tokenlowcost.com/v1",
  apiKey: "sk-tlc-your-key",
});

const r = await client.chat.completions.create({
  model: "claude-opus-5.5",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(r.choices[0].message.content);

The TypeScript examples on this page are ES modules: save one as main.mts and run npx tsx main.mts.

Streaming

Add stream=True. The token counts come in the last chunk, which can arrive without choices, so check chunk.choices first.

stream = client.chat.completions.create(
    model="claude-opus-5.5",
    messages=[{"role": "user", "content": "Write a haiku about caching"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="")

The Responses API works too: client.responses.create(...).

Use the same base URL as everywhere else. The SDK adds /v1/messages to it, and that's fine. Some models send a thinking block before the text, so the example prints only the text blocks.

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.tokenlowcost.com/v1",
    api_key="sk-tlc-your-key",
)

msg = client.messages.create(
    model="claude-opus-5.5",
    max_tokens=512,
    messages=[{"role": "user", "content": "Hello"}],
)
for block in msg.content:
    if block.type == "text":
        print(block.text)

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://api.tokenlowcost.com/v1",
  apiKey: "sk-tlc-your-key",
});

const msg = await client.messages.create({
  model: "claude-opus-5.5",
  max_tokens: 512,
  messages: [{ role: "user", content: "Hello" }],
});
for (const block of msg.content) {
  if (block.type === "text") console.log(block.text);
}

The same key works in all three request formats.

curl https://api.tokenlowcost.com/v1/chat/completions \
  -H "Authorization: Bearer sk-tlc-your-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-opus-5.5", "messages": [{"role": "user", "content": "Hello"}]}'

curl https://api.tokenlowcost.com/v1/messages \
  -H "x-api-key: sk-tlc-your-key" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-opus-5.5", "max_tokens": 256, "messages": [{"role": "user", "content": "Hello"}]}'

curl https://api.tokenlowcost.com/v1/responses \
  -H "Authorization: Bearer sk-tlc-your-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6-sol", "input": "Hello"}'

Check your balance and your last 20 requests:

curl https://api.tokenlowcost.com/v1/me -H "Authorization: Bearer sk-tlc-your-key"

When something goes wrong, you get a status code and a short message.

Spending limits and expiry dates are set per key in the dashboard. Your balance is under Billing.