OpenAI

Anthropic

Gemini

DeepSeek

Mistral

One API key.Every major LLM.One bill.

Drop-in replacement for the OpenAI SDK. One key for every model with transparent per-token billing — you set the preference, we deterministically pick the model and transparently handle context overflow and provider failures. Built for AI agents (Claude Code, Codex, OpenClaw…) to wire in with three commands.

Get started See the code

agent — tokenroute

How it works

Three things we handle for you.

OpenAI-compatible

Same SDK, same shape. Point base_url and swap the key.

5 providers, one key

OpenAI, Anthropic, Gemini, DeepSeek, Mistral — single bill, single dashboard.

Prepaid, per-token

No subscriptions, no minimums. Top up, watch cost in real time.

Live demo · LIVE

What model: "auto" actually does

No difficulty guessing — you set the preference, we pick deterministically and handle the rest. Pick a preference to see the model:

Cost

Balanced

Quality

Long input ↑

tokenroute /v1

model: "auto"

Cost

Cheapest · low latency

Balanced

Best speed / intelligence balance

claude-sonnet-4-6

$3.00 / $15.00 / 1M

Quality

Top-tier reasoning / agentic

Long input ↑

🛡 guardrail

Input > 180k auto-escalates to a large-context model

* In production the gateway picks the model deterministically from your preference and auto-handles context overflow / provider failures (not difficulty guessing). Prices are estimates per million tokens.

Quick start

Five lines of Python.

quickstart.py

from openai import OpenAI
 
client = OpenAI(
    api_key="sk-tr-…",                    # from /dashboard/keys
    base_url="https://api.tokenroute.io/v1",
)
 
response = client.chat.completions.create(
    model="claude-sonnet-4-5",            # or gpt-4o, gemini-2.5-pro, deepseek-chat…
    messages=[{"role": "user", "content": "Summarize this commit log."}],
)
print(response.choices[0].message.content)

Any OpenAI-compatible client works — Node.js, Go, Ruby, curl. Streaming, tool calls, JSON mode, vision — all passed through unchanged.

Built for AI agents

Your agent ships LLM apps in five commands.

tokenroute exposes a CLI and (soon) a remote MCP server so Claude Code, Codex, OpenClaw, Hermes and other coding agents can install, configure, test, and even request top-up — without any web dashboard clicks. Every command supports `--json` and respects `TOKENROUTE_API_KEY` env var for non-interactive sub-agent runs.

agent.sh

# install + login (one-time per user)
$ uvx tokenroute login                            # opens browser
 
# wire up a project
$ TOKENROUTE_JSON=1 uvx tokenroute keys create --name myapp   # raw key returned once
$ uvx tokenroute env >> .env                      # OPENAI_API_KEY + BASE_URL
$ uvx tokenroute test                             # verify it works
 
# need credit? agent surfaces the URL — does NOT auto-pay
$ TOKENROUTE_JSON=1 uvx tokenroute topup --amount 5
{"checkout_url":"https://checkout.stripe.com/c/pay/..."}

Hard constraint

Agents must NOT auto-pay. `topup` returns a Stripe Checkout URL — the agent hands it to the user. This is a KYC red line we won't cross.

Read the agent guide →

Pricing

Pay per token. No subscriptions.

We bill upstream provider list price + 30% routing fee. No monthly fee, no minimums. Balance never expires. Per-token prices below in USD per million input / output tokens:

Model	Provider list	Tokenroute
GPT-4o	$2.50 / $10.00	$3.25 / $13.00
Claude Sonnet 4.5	$3.00 / $15.00	$3.90 / $19.50
Gemini 2.5 Pro	$1.25 / $10.00	$1.63 / $13.00
DeepSeek-V3	$0.27 / $1.10	$0.35 / $1.43

Full price list at dashboard/models after sign-up.Create an account

One API key.Every major LLM.One bill.

from openai import OpenAI

client = OpenAI(

api_key="sk-tr-…", # from /dashboard/keys

base_url="https://api.tokenroute.io/v1",

)

response = client.chat.completions.create(

model="claude-sonnet-4-5", # or gpt-4o, gemini-2.5-pro, deepseek-chat…

messages=[{"role": "user", "content": "Summarize this commit log."}],

)

print(response.choices[0].message.content)

# install + login (one-time per user)

$ uvx tokenroute login # opens browser

# wire up a project

$ TOKENROUTE_JSON=1 uvx tokenroute keys create --name myapp # raw key returned once

$ uvx tokenroute env >> .env # OPENAI_API_KEY + BASE_URL

$ uvx tokenroute test # verify it works

# need credit? agent surfaces the URL — does NOT auto-pay

$ TOKENROUTE_JSON=1 uvx tokenroute topup --amount 5

{"checkout_url":"https://checkout.stripe.com/c/pay/..."}

Model

Provider list

Tokenroute

GPT-4o

$2.50 / $10.00

$3.25 / $13.00

Claude Sonnet 4.5

$3.00 / $15.00

$3.90 / $19.50

Gemini 2.5 Pro

$1.25 / $10.00

$1.63 / $13.00

DeepSeek-V3

$0.27 / $1.10

$0.35 / $1.43