TokenLayer API

Introduction

TokenLayer is a unified inference gateway. You authenticate every request with a single TokenLayer API key and can reach models from OpenAI, Anthropic, Google, DeepSeek, Meta, and more — all through one endpoint, with one balance and one usage dashboard.

The API speaks two dialects, so whichever ecosystem your code already lives in keeps working:

Each model is served on the API dialects it declares — most are OpenAI-only, Claude models speak both. The model picker below and GET /v1/models (the supported_endpoint_types field) show which dialects every model supports, and our code samples only ever pair a model with an endpoint it actually serves.

Quickstart

1. 1Create an account and open the API Keys page in your dashboard.
2. 2Click Generate key. The full key (sk-…) is shown once — store it somewhere safe.
3. 3Send your first request:

curl https://api.tokenlayer.net/v1/chat/completions \
  -H "Authorization: Bearer YOUR_TOKENLAYER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4",
    "messages": [{ "role": "user", "content": "Say hello!" }]
  }'

That's it. The response is identical in shape to what the upstream provider would return, so existing parsers, SDKs, and frameworks work unchanged.

Authentication

Every request must carry your TokenLayer API key. Two header styles are accepted on all endpoints:

Keys are created and revoked from the dashboard. Revocation takes effect immediately. All keys on your account draw from the same credit balance.

OpenAI-compatible API

A faithful implementation of the OpenAI API. Use the official openai SDK in any language — just change the base URL and the key.

curl https://api.tokenlayer.net/v1/chat/completions \
  -H "Authorization: Bearer YOUR_TOKENLAYER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4",
    "messages": [
      { "role": "system", "content": "You are a concise assistant." },
      { "role": "user", "content": "Explain HTTP caching in two sentences." }
    ],
    "max_tokens": 200,
    "temperature": 0.7
  }'

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1718000000,
  "model": "gpt-5.4",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "HTTP caching stores copies of responses..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 28,
    "completion_tokens": 52,
    "total_tokens": 80
  }
}

curl https://api.tokenlayer.net/v1/chat/completions \
  -H "Authorization: Bearer YOUR_TOKENLAYER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4",
    "messages": [{ "role": "user", "content": "Write a haiku." }],
    "stream": true
  }'

# data: {"choices":[{"delta":{"content":"Tok"}}], ...}
# data: {"choices":[{"delta":{"content":"ens"}}], ...}
# data: [DONE]

curl https://api.tokenlayer.net/v1/models -H "Authorization: Bearer YOUR_TOKENLAYER_KEY"

Anthropic-compatible API

If your code is built on the Anthropic SDK or the Messages API format, point it at TokenLayer instead — authentication, request shape, and streaming all match.

curl https://api.tokenlayer.net/v1/messages \
  -H "x-api-key: YOUR_TOKENLAYER_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4-6",
    "max_tokens": 1024,
    "system": "You are a concise assistant.",
    "messages": [
      { "role": "user", "content": "Explain HTTP caching in two sentences." }
    ]
  }'

{
  "id": "msg_abc123",
  "type": "message",
  "role": "assistant",
  "model": "claude-opus-4-6",
  "content": [
    {
      "type": "text",
      "text": "HTTP caching stores copies of responses..."
    }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 31,
    "output_tokens": 47
  }
}

Using official SDKs

There's no TokenLayer-specific SDK to install — use the official OpenAI or Anthropic libraries and override two settings:

The Anthropic SDKs append /v1/messages to their base URL themselves, which is why their base URL stops at /api.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.tokenlayer.net/v1",
    api_key="YOUR_TOKENLAYER_KEY",
)

Coding tools

AI coding tools are the most common way to use TokenLayer. Each of these speaks one of our two API dialects, so setup is just pointing the tool at our base URL with your key.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.tokenlayer.net",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_TOKENLAYER_KEY",
    "ANTHROPIC_MODEL": "claude-opus-4-6"
  }
}

export ANTHROPIC_BASE_URL="https://api.tokenlayer.net"
export ANTHROPIC_AUTH_TOKEN="YOUR_TOKENLAYER_KEY"
export ANTHROPIC_MODEL="claude-opus-4-6"
claude

# Use TokenLayer as the model provider
model = "gpt-5.4"
model_provider = "tokenlayer"

[model_providers.tokenlayer]
name = "TokenLayer"
base_url = "https://api.tokenlayer.net/v1"
env_key = "TOKENLAYER_API_KEY"
wire_api = "responses"

export TOKENLAYER_API_KEY="YOUR_TOKENLAYER_KEY"
codex

Errors

Errors use conventional HTTP status codes with a JSON body: { "error": { "message": "...", "type": "..." } }.

{
  "error": {
    "message": "Insufficient credits. Top up to continue.",
    "type": "insufficient_quota"
  }
}

Credits & limits

TokenLayer is prepaid. Usage is metered per request against your account's credit balance at the per-model rates shown on the pricing page. All of your API keys share one balance; per-key spend is broken out on the API Keys page so you can see which key is consuming what.

• When the balance reaches zero, requests return 402 until you top up — nothing is queued or silently dropped.
• Live usage, token counts, and spend-per-model are on the Usage page.
• Top up from the Topup page; credits never expire.