OmniRoute is a free, open-source tool that connects hundreds of AI providers to a single address on your own computer. You plug in the free tiers those providers already offer, and OmniRoute gives every app on your machine one endpoint to talk to. When one provider runs out of quota, it hands off to the next automatically, so you rarely hit a wall.
This guide walks you through the whole thing on Windows, from a blank machine to a working setup wired into VS Code. You don’t need any terminal experience. Every command is here to copy, and the spots where beginners usually get stuck are called out as you go.
Everything here uses the official free tiers that providers publish themselves. You sign in with your own account and stay inside their rules. Nothing in this guide gets around a paywall.
If you’d rather watch, the full video walkthrough is here:
What you’ll find in this guide:
– What OmniRoute is
– How the free tiers work
– What you need before you start
– Step 1: Install Node.js
– Step 2: Install OmniRoute
– Step 3: Start the server and open the dashboard
– Step 4: Connect your first free model
– Step 5: The four ways to connect a provider
– Step 6: Test your providers
– Step 7: Try it in the Playground
– Step 8: Set up a Combo so you never get stuck
– Step 9: Turn on compression to stretch your quota
– Step 10: Connect OmniRoute to VS Code
– Starting and stopping OmniRoute
– Troubleshooting
– Commands at a glance
– Free provider links
– FAQ
What OmniRoute is
Think of OmniRoute as a power strip for AI models. Normally, every tool you use wants its own connection: your code editor wants one key, your chat app wants another, and every provider has its own login and its own rate limit. OmniRoute sits in the middle. You connect your providers to OmniRoute once, and everything else connects to OmniRoute. One address, one key.
A few things make it worth the setup:
It’s self-hosted. It runs on your computer as a local proxy. Your keys stay on your disk, encrypted, and your prompts go to the provider you picked and nowhere else.
It has automatic fallback. When a provider hits its limit, OmniRoute moves to the next one in milliseconds. You don’t have to do anything.
It’s open source under the MIT license. You can read the code, and there’s no billing system anywhere in it. The dollar figure on the dashboard is a savings counter, not a bill.
How the free tiers work:
Most major AI providers publish a free tier on their own pricing pages. NVIDIA, Cerebras, Cloudflare, Groq, Google, and a long list of others all offer one. You sign up with your own account, generate a key or sign in, and use your own quota inside each provider’s rules.
OmniRoute doesn’t change any of that. It collects those free tiers in one place so you’re not juggling a dozen dashboards and a dozen keys. Read each provider’s terms before you rely on one, and check the current limits on their page, because free tiers change over time.
What you need before you start:
– A Windows 10 or 11 machine.
– Node.js, version 22.x or 24.x. This is the one prerequisite, and it’s covered in Step 1.
That’s the whole list. The rest installs in a couple of commands.
Step 1: Install Node.js
Node.js is the engine OmniRoute runs on. This is where most people get stuck, so take it slowly.
1. Go to [nodejs.org](https://nodejs.org).
2. Click Get Node.js, then on the next screen scroll down and click the Windows Installer (.msi) download.
3. Run the installer and click through it with the default settings. Next, Next, Install.
Once it finishes, open Command Prompt to check the version. Press the Windows key, type `cmd`, and open Command Prompt.
Use Command Prompt, not PowerShell. A couple of small differences trip beginners up, and Command Prompt behaves.
In the black window, type:
node -v
Press Enter. You want a version number that starts with 22 or 24.
This part is specific: Node 21 and 23 are not supported. It’s a database library OmniRoute uses under the hood, so if you land on 23, it will break with no obvious reason. If your version doesn’t start with 22 or 24, reinstall from nodejs.org and grab the LTS build.
Step 2: Install OmniRoute
In the same Command Prompt window, type:
npm install -g omniroute
Press Enter. The `-g` installs it globally so you can run it from anywhere. This takes a minute or two, and you’ll see a wall of text scroll past. That’s normal. Let it finish until your cursor comes back.
If you get a red permission error: close the window, open the Start menu, type `cmd`, then right-click Command Prompt and choose Run as administrator. Run the same command again. That clears it almost every time.
Step 3: Start the server and open the dashboard
Still in Command Prompt, type one word:
omniroute
Press Enter. OmniRoute boots and prints a localhost link in the window telling you the dashboard is live.
Copy that link straight from Command Prompt and open it in your browser. The default address is:
http://localhost:20128
If it asks for a password on first login, type `CHANGEME` in all caps and log in.
The dashboard loads, and it’s running entirely on your computer. Down the left is the sidebar: Endpoints, API Keys, Providers, Combos, Analytics, Health, and Settings. If your sidebar looks shorter, there’s a preset toggle at the top that switches between a minimal view and a fuller one.
Keep the Command Prompt window open. That window is the server. If you close it, OmniRoute stops. Minimize it and leave it running while you work.
Step 4: Connect your first free model:
Click Providers in the sidebar. The catalog is large, with hundreds of providers, and many of them have a free tier. The count changes with almost every release, so whatever number is on your screen is the current one.
For your first connection, pick the easiest path. Search for OpenCode Free and click Connect. This one needs no sign-in at all, so you get a working free model in about five seconds. The card flips to a connected state, and you’re routing through your own machine.
Step 5: The four ways to connect a provider:
Providers aren’t all connected the same way, and picking the right kind saves you trouble later. OmniRoute sorts them by how you log in, and you’ll see four types.
1. No sign-in. Some providers just work with nothing to enter. OpenCode, which you just added, is one. Good for testing.
2. API key. You generate a key on the provider’s own site and paste it in. DeepSeek works this way, along with NVIDIA, Groq, Cerebras, Mistral, and OpenRouter.
3. App sign-in. You click Connect, a login window opens, and you approve it. Claude, GitHub, and Codex sit here. It’s convenient, but the login can expire and need reconnecting.
4. Browser cookies. A couple of providers let you paste cookies from a logged-in browser tab. It works, but it’s the fragile option, because cookies expire quickly.
The recommendation: lean on the API-key providers as your backbone. A key is stable, it maps to a clear free-tier limit, and it doesn’t break the way a login or a cookie does.
Here’s the pattern for an API-key provider, using DeepSeek as the example:
1. In OmniRoute, click the DeepSeek card and choose Add API Key.
2. Open the provider’s own site in a new tab, sign up, find the API keys page, and create a key.
3. Copy the key, paste it back into OmniRoute, and save.
OmniRoute runs a quick test, the card turns green, and the provider’s models appear. It’s the same three moves for every API-key provider, so connect two or three while you’re here. That stack matters in the Combos step.
Step 6: Test your providers:
Once you have a few providers connected, confirm they’re actually live. Open the card for one of them and click Test All Models. OmniRoute runs a quick test against the key you gave it and checks which models respond.
Then go back to the Home page and look at the provider topology in the middle of the screen. You’ll see animated dotted lines running between your providers. That’s the signal that your connections are up and routing.
Step 7: Try it in the Playground:
The quickest way to talk to your models is right in the dashboard. Open the Playground.
For the model name, type one word:
auto
That tells OmniRoute to look at everything you’ve connected and pick for you. Type a message and send it. The response streams back, and it shows you which provider actually answered.
If you prefer the terminal, open a second Command Prompt window (the first one is still running the server) and type `omniroute chat` for the same thing in your terminal.
Step 8: Set up a Combo so you never get stuck:
Free tiers run out. A Combo is your backup plan: a list of models in order, so when the first runs dry, OmniRoute quietly moves to the second, then the third. You do nothing.
There are two ways to set this up.
The easy way is to use `auto` as your model name, which you already did in the Playground. OmniRoute scores everything you’ve connected and picks live.
For full control, open Combos in the sidebar and build your own. Pick your models, drag them into the order you want, and choose a strategy. For your first one, use priority, which means go top to bottom.
Underneath all of this, three safety nets run automatically. If a whole provider goes down, a circuit breaker skips it until it recovers. If one key gets rate-limited, that key cools off while your others keep serving. If one model hits its cap, it locks out that single model and everything else keeps running. The result is a handoff you never notice instead of your AI stopping.
Step 9: Turn on compression to stretch your quota:
When you work with AI, especially inside a code editor, you send it a lot of repeated context: log files, command output, the same background over and over. All of that burns tokens. Compression shrinks it before it leaves your machine.
It only touches what you send in, not the answer coming back, and it protects code blocks, links, and structured data so they pass through untouched.
Open the compression settings, and you’ll see presets that trade a little accuracy for a lot of saved tokens. Start on a lighter preset, check that your answers still look right, then push it higher. Cutting your usage in half means you get twice the work out of the same free tier.
Step 10: Connect OmniRoute to VS Code:
This is where it pays off. You can connect OmniRoute to the chat that’s built into VS Code, with no extra extension. VS Code is free and open source, so anyone can follow along.
1. Open VS Code. In the top right, just inside the search bar, click the Toggle Chat icon. A chat panel opens on the right.
2. In the chat, click where it says Auto. A small menu appears. Depending on your VS Code version, you may see a gear icon here. If you do, click it. If not, click Manage Models, then Add Models.
3. In the dropdown, pick Custom Endpoint.
4. It asks for the endpoint. In OmniRoute, open the Endpoints page, copy the endpoint shown there, paste it into VS Code, and press Enter.
5. It asks for an API key. In OmniRoute, open the API Keys page, create a key (name it anything, keep the defaults), copy it, paste it into VS Code, and press Enter.
6. It asks for the request and response format. Choose Chat Completions.
7. It opens a small model config file. You fill in three things: the model ID, a name, and the URL. The model ID comes from your Providers page in OmniRoute. The name is whatever you want. The URL is the same endpoint you copied in step 4.
For example, to add DeepSeek, copy its model ID from the Providers page, paste it into the config, and give it a name like `OR-deepseek` so it’s easy to spot. Save the file.
Now click Auto in the chat again, and your new model appears in the list. Select it and start chatting. To add another as a fallback, repeat the same steps with a second provider, like Google Gemini.
That’s the whole point of this setup: instead of bouncing between a dozen sites, all your free models live in one editor. Configure a few, and it’s hard to run out, because the limits reset and one hands off to another when it’s spent.
Commands at a glance:
node -v Check Node version (needs 22.x or 24.x)
npm install -g omniroute Install OmniRoute
omniroute Start the server
omniroute chat Chat from the terminal
Free provider links:
Each of these has its own free tier and its own rules. Create an account, generate a key, and read the terms before you rely on one.
– OpenCode Free: no sign-up, works inside OmniRoute
– DeepSeek: [platform.deepseek.com](https://platform.deepseek.com)
– Google AI Studio (Gemini): [aistudio.google.com](https://aistudio.google.com)
– Groq: [console.groq.com](https://console.groq.com)
– Cerebras: [cloud.cerebras.ai](https://cloud.cerebras.ai)
– NVIDIA: [build.nvidia.com](https://build.nvidia.com)
– Mistral: [console.mistral.ai](https://console.mistral.ai)
– OpenRouter: [openrouter.ai](https://openrouter.ai)
OmniRoute itself: [github.com/diegosouzapw/OmniRoute](https://github.com/diegosouzapw/OmniRoute) and [omniroute.online](https://omniroute.online).
FAQ:
Is OmniRoute actually free?
Yes. It’s open source under the MIT license, it has no billing system, and it runs entirely on your machine. The AI models you connect use the providers’ own free tiers.
Is it safe to put my API keys in?
Your keys are stored on your own disk, encrypted, and OmniRoute runs locally as a proxy. Your prompts go to the provider you chose and nowhere else. Because it’s open source, you can inspect the code yourself.
Does this get around paying for AI?
No. It uses the free tiers providers publish on their own pricing pages, with your own account, inside their rules. It just collects them in one place.
Which models can I use?
Hundreds of providers are supported, covering major text models and, in newer versions, image and video models too. The exact list is on the Providers page and grows with each release.
Does it only work on Windows?
This guide is written for Windows, but OmniRoute also runs on macOS, Linux, and Docker. The commands are the same across platforms.
Do I have to set it up again after restarting my computer?
No. Your configuration is saved on disk. After a restart, you only need to start the server again by running `omniroute` in Command Prompt.
Will I really not run out of tokens?
It’s hard to run out if you connect several free tiers, because the limits reset and OmniRoute falls back to another provider when one is spent. Heavy use can still hit limits, which is exactly what Combos and compression are for.
Found this useful? The full video walkthrough covers every step on screen. If you get stuck, run `omniroute doctor` and share the output.
