Discord Download

OpenLLM vs AutoRouter

If you’re looking at OpenLLM to use your plans in all your coding agents, AutoRouter does that with no keys, daemon or fallback chains to set up. It’s a Mac menu-bar app that pools the AI subscriptions you already pay for, picks the model for each turn and adds itself to each app when you click Enable. OpenLLM is a hosted gateway with a daemon on each of your machines: you get one key and endpoint, and choose the model or chain each tool asks for. This page covers how OpenLLM works, then how the two compare. Learn more

AutoRouter A Mac app that picks a model for each turn. $0/mo on your plans Download OpenLLM A hosted gateway with a daemon on your machines. $0/mo, Pro $30
Price Free to start on your plans Free with 50M tokens a month and 2 devices, Pro $30 a month with unlimited tokens and 5 devices
Fee on tokens Provider’s price plus 10% on paid models None listed on its pricing page
Runs on Your Mac, macOS 14 or later Its hosted gateway, plus a daemon on macOS or Linux and a Mac app
Tools Codex, Claude Code, Cursor, Claude Desktop, T3 and Terminal Any OpenAI or Anthropic compatible tool, with Claude Code, Codex, opencode, Goose, Raycast, n8n and more named
Setup Install, sign in, click Enable Install, set up a key vault, sign in to each vendor CLI, point each tool at a base URL and key
Your plans Claude, ChatGPT, Grok and Gemini plans, in any harness Claude, ChatGPT, Kimi and Grok in its docs, plus Cursor and opencode Go on its site, API keys and local models
Model choice Your rules, then a model that clears the quality bar The model or alias each tool asks for, then the next in its fallback chain on a limit or error
Each request Every turn shows its model, cost and why Usage and cost analytics in its dashboard
Requests go Through AutoRouter’s servers to each provider Subscriptions through your daemon, keys and included models through its gateway
  • A model picked for each turn, not a chain
  • No keys, base URLs or daemon to point tools at
  • Your plans stacked and used first
  • Rollout to every Mac in a team through MDM

Sources checked November 19, 2026: OpenLLM home page (Mac app version 2.10.22), pricing, docs overview, install docs, subscriptions docs, routing docs, client setup docs and AutoRouter pricing. AutoRouter wrote this page. See every comparison.

How OpenLLM works

What the product is, how setup works, and how it chooses where a request runs.

One key for your plans, keys and local models

OpenLLM describes itself as “one key for every AI you pay for” (home page). It gives your tools one OpenAI and Anthropic compatible endpoint and one sk-llm-… key. Behind it sit a hosted gateway, which serves API-key and included models and runs the dashboard, and a daemon on each Mac or Linux machine, which serves your subscriptions and local model servers such as Ollama and LM Studio (docs). Every account comes with included inference, so requests work before you connect anything.

How setup works

Install the CLI and daemon with one line, or download the Mac app, which runs the same installer (install docs):

curl -fsSL https://www.openllm.sh/install | bash

Then connect each subscription: the daemon needs the vendor’s own CLI on that machine and keeps its own login, so you sign in again through openllm auth login <provider> (subscriptions docs). Point each tool at OpenLLM with its base URL and your key, for example ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY for Anthropic compatible tools, or start one through the CLI, such as openllm claude.

Fallback chains

A fallback chain is an ordered list of up to 16 models under one alias. A tool that asks for the alias gets the first model, and OpenLLM moves to the next on rate limits, quota or auth failures and some server errors, as long as nothing has been sent back yet. A model that hits a limit is skipped for 3 minutes, and a request too large for a model can hop to one with a bigger context window. A tool that asks for a concrete model ID gets only that model (routing docs).

Devices and keys

Agents on other machines can reach the same plans with a URL and an OpenLLM key. Through the hosted gateway, a subscription model runs only when one of your devices is online, and the docs say OpenLLM’s servers never receive the subscription token. Provider keys are encrypted in the browser under a recovery phrase only you hold.

What it does well

  • One endpoint takes subscriptions, API keys, local models and included models together.
  • It works with any tool that speaks the OpenAI or Anthropic API, including your own scripts and n8n flows.
  • Fallback chains keep a request going when one plan or key hits its limit.
  • The daemon runs on Linux as well as macOS, and agents on other machines can share it.
  • Free includes 50M tokens a month, so it works before you connect anything.
  • Extras include memory shared across agents, an MCP server, browser chat and a “Connect with OpenLLM” button for app developers.

Where it gets harder

  • You choose the model. OpenLLM serves the model or chain a tool asks for, and moves on when a request fails, not when a task is harder or easier.
  • Each tool needs its own base URL and key, or has to be started through the openllm CLI.
  • Each subscription needs the vendor’s CLI and a separate sign-in on the device that serves it, and that device has to be online.
  • Free connects 2 devices and Pro 5.
  • There’s no Windows build of the daemon.

How AutoRouter differs

OpenLLM gives your tools a key and moves down a chain when a request fails. AutoRouter chooses the model for every turn, using models it includes and the plans you already pay for.

It picks the model for each turn

AutoRouter shows up as one more model in the picker you already use. While it’s selected, it works out how hard each turn is and only considers models that clear the quality bar for it. Hard work goes to the strongest model your plans cover, and quick work to the one that costs you least. Your own plain-English rules go first, such as “Prefer GPT-6.1 Sol for code reviews.” Every turn shows its model, what it cost and why.

Nothing to point your tools at

There’s no key to mint, base URL to set or daemon to keep online. You sign in, sign in to your plans and click Enable for each app. Enabling adds AutoRouter to Codex’s config.toml and Claude Code’s settings.json after saving a copy of each file, and disabling it switches the app back. Frontier and open-weight models come with AutoRouter, kept current by its team.

Your plans, stacked in any harness

You can connect the AI subscriptions you already pay for and use them in any harness, such as a ChatGPT plan in Claude Code. When one account reaches its limit, AutoRouter moves to the one with the most room left before it spends any credit. Free connects two subscriptions and Pro connects any number.

Built for teams to roll out

Enterprise rolls AutoRouter out to every Mac through Jamf or another MDM, with nothing for anyone to set up. Admins get allowed models, team budgets and zero data retention in one place, spend by team, project or ticket, and SSO, SCIM and an audit log.

What it costs and what it can’t do

Routing across your plans is free, and paid models bill at the provider’s price plus 10% (pricing). It runs only on macOS 14 or later, and it doesn’t give you a raw API endpoint for your own scripts and SDKs the way OpenLLM does. It sometimes misjudges how hard a task is. Requests pass through AutoRouter’s servers on their way to each provider; see privacy for what is kept.

Which one to choose

Choose OpenLLM if you want one key for scripts, apps and agents across Mac and Linux machines, want to mix local models and API keys with your plans, or want a fallback chain you order yourself.

Choose AutoRouter if you’re on a Mac and want each turn in every coding agent you use sent to the right model, with your plans used first and nothing to point your tools at. For other options, read Model routers for Claude Code and Codex, compared.

Setup, side by side

The steps each one’s own docs give, as of November 19, 2026.

OpenLLM

  1. Install Run the one-line installer, or download the Mac app, and sign in in the browser.
  2. Set up your vault Save the recovery phrase that encrypts your provider keys.
  3. Connect subscriptions Run openllm auth login <provider> on the device that serves each plan.
  4. Point your tools Set each tool’s base URL and sk-llm-… key, or start it through the openllm CLI.

AutoRouter

  1. Download AutoRouter for macOS, signed and notarized by Apple.
  2. Sign in Sign in or create a free account, then sign in to your plans.
  3. Enable your apps Allow Accessibility and click Enable for each app.
  4. Pick AutoRouter In a new session, choose AutoRouter from Codex’s model menu, or with /model in Claude Code.

Questions

What people ask about OpenLLM.

What is OpenLLM?

OpenLLM, at openllm.sh, is a routing layer that gives your tools one OpenAI and Anthropic compatible endpoint and one key. A hosted gateway serves API-key and included models, and a daemon on your Mac or Linux machines serves your subscriptions and local models.

How much does OpenLLM cost?

Its pricing page lists Free at $0 with 50M tokens a month and 2 devices, and Pro at $30 a month with unlimited tokens, 5 devices and a 3-day free trial.

Which subscriptions does OpenLLM support?

Its subscriptions docs list Claude Pro or Max, a ChatGPT subscription, a Kimi Code plan, and SuperGrok or X Premium+. Its home page also names Cursor and opencode Go. AutoRouter connects Claude, ChatGPT, Grok and Gemini plans.

Is OpenLLM an alternative to Claude Code Router?

Both put other models behind Claude Code. Claude Code Router is a free proxy you configure with API keys and routing rules; OpenLLM is a hosted gateway and daemon that also serves your subscriptions. AutoRouter is a third option: a Mac app that picks the model for each turn. See Claude Code Router vs AutoRouter.

Does OpenLLM pick a model for me?

It runs the model or alias each tool asks for. An alias is a fallback chain you order, and OpenLLM moves to the next model on a limit or error, or to a bigger context window when a request is too large. AutoRouter picks a model for each turn by how hard it is, after your own rules.

Do my requests go through OpenLLM’s servers?

Its docs say subscription requests run through the daemon on your device and that its servers never receive the subscription token, while API-key and included models are served by its hosted gateway. AutoRouter’s requests pass through its servers to each provider; see privacy.

How is AutoRouter different from OpenLLM?

OpenLLM is one key and endpoint for any tool, with fallback chains you order, on Mac and Linux. AutoRouter is a Mac app with models included and no keys, picks a model for each turn with its reasons and cost, uses your plans first, and rolls out to a team’s Macs through MDM.

What does AutoRouter cost?

Routing across your plans is free, and paid models bill at the provider’s price plus 10%. See pricing, or estimate a month with and without AutoRouter in the savings calculator.

Your plans, every turn routed

Install the Mac app, sign in to your plans and click Enable.

Download AutoRouter for Mac