xcb
Get started
Theme
Appearance

How xcb compares

xcb vs OpenRouter

OpenRouter is a hosted API that sends each request from your code to one of hundreds of models and bills you per token. xcb runs on your machine and sends each coding task to one of the Claude, Codex, or Devin accounts you already pay for. Pick OpenRouter to call many models from an app; pick xcb to spread coding work across subscriptions you already have.

Native xcb is a source preview for macOS and Linux.

What each product routes

Both call themselves routers. OpenRouter chooses a model and provider for each API request; xcb chooses one of your accounts for each coding task.

OpenRouter: one API key for hundreds of models

Your code posts to OpenRouter’s OpenAI-compatible endpoint with one API key, and OpenRouter decides which provider serves the model you named, or picks a model for you with openrouter/auto. Its June 2026 routing guide describes 400+ models across 70+ providers, provider ordering by price, fallback chains across models and providers, and :nitro and :floor variants for throughput and price.

Billing is pay-as-you-go per token in credits at the provider’s list price, with rate-limited :free models for trying the service. A BYOK path can send requests under your own provider keys instead.

xcb: one of your accounts for each task

You sign in to your Claude, Codex, or Devin accounts on your own machine. For each task, xcb picks an account that is signed in, idle, and outside any known quota window, on a model it has recently seen in that provider’s catalog, and holds that account until the provider process exits.

The provider’s own runtime runs the task under your subscription, so xcb has no token billing and no model catalog of its own, and it never proxies or rewrites API traffic. An unknown model name cannot activate a provider. Another agent can hand xcb a task with xcb --json route, and applications can embed the TypeScript SDK.

Side by side

Updated . OpenRouter details come from its site, pricing page, and routing guide on that date, and its catalog, routing options, and prices change. xcb details come from its own documentation.

On a small screen, scroll the table sideways to compare each approach.

OpenRouter and xcb at a glance
AspectOpenRouterxcb
What it routes

One API request: a prompt in, a model’s answer out

One coding task; each route call runs one provider turn on one model

What you bring

An OpenRouter API key, or your own provider keys through its BYOK option

Claude, Codex, or Devin subscriptions you already pay for, signed in locally

Where it runs

OpenRouter’s hosted service; requests leave your machine for its endpoint

Your Mac or Linux machine; the provider’s own runtime makes the model calls

How you pay

Per token from a prepaid credit balance, at the provider’s list price; some :free models with daily rate limits

xcb charges nothing; usage draws on each subscription’s own allowance

Choosing the model

The model field on each request, or the openrouter/auto router

xcb picks from models recently seen on accounts that can take the task now, ranked by task type and by relative quality, cost, and latency

Choosing the provider

The provider object sets provider order, price ceiling, and region limits; price-ordered by default

The provider of the chosen account; a task stays on one account, so its requests are not spread across providers

When a call fails

Fallback chains retry on another model or provider automatically

The task stays on its account; when xcb cannot confirm how a run ended, it keeps the account held and does not retry

OpenRouter also serves image, video, and speech models, which this table leaves out.

Which one to pick

Pick OpenRouter to reach many models from your own code

You want one key and one bill across many models, automatic fallbacks when a provider rate-limits or errors, or a model your subscriptions do not include. Any app that can call the OpenAI chat API can use it.

Pick xcb to spread coding tasks across plans you pay for

You already pay for Claude, Codex, or Devin and want each coding task sent to an idle account and held there until its process exits, with no per-token charge. Your account credentials stay out of your project folder.

Compare cost on your own workload

Per-token credits and a subscription’s usage allowance measure different things, so which one costs less depends on how you work.

Back to all comparisons →