Wend router / Free open-source LLM router

Updated 10 November 2026

open source GitHub Guide

Wend router is a free, open-source LLM router: one local address, compatible with the OpenAI API and the Anthropic API, in front of many LLM providers. Your agents talk to one address; the router sends each request to a model that is up and fits the job.

When a provider fails, runs out of quota or stalls, Wend moves the request to another model on its own, so the agent does not stop. Free tiers come first; paid models are used only when you allow them, within a daily budget you set.

It is one program for Linux, macOS and Windows, with a dashboard built in. Some providers work without any key, so the router answers before you add a single one. Keys you add are stored encrypted on your own machine.

Quick start

Linux or macOS:

curl -fsSL https://raw.githubusercontent.com/dklymentiev/wend-router/main/install.sh | sh

Windows, in PowerShell:

irm https://raw.githubusercontent.com/dklymentiev/wend-router/main/install.ps1 | iex

No root or administrator rights. The script starts the router, sends one test request and opens the dashboard at http://127.0.0.1:8888/dashboard. It runs on a laptop or on your own server.

Profiles: one name per kind of job

The model name an agent sends can be a profile. The router then picks only among the models of that profile and switches between them.

tools: models that call tools well, for coding agents such as Claude Code.
no-think: models without a reasoning phase, quickest first.
vision: models that read images.
auto: every model, strongest first. fast: quickest to start first.
private: only providers that do not train on prompts.
Your own profile from the dashboard, for example think for reasoning models.

Connect your agents

Claude Code:

export ANTHROPIC_BASE_URL=http://localhost:8888
export ANTHROPIC_AUTH_TOKEN=wend
export ANTHROPIC_MODEL=tools
claude

Any tool that takes an OpenAI address uses base URL http://localhost:8888/v1, any text as the API key, and a profile as the model. Flint Agent has a ready tab in the router's Connect section.

Step-by-step guide

How to run Claude Code for free walks through the setup and connects four agents to four profiles. The same guide is available as a free 10-page PDF on that page.

Honest limits

Free models are weaker than the best paid models. For production work, where real users or customer data depend on the answers, use paid models; the router can stay in front of them within a budget.
Free tiers have daily limits per provider. The router moves to another model when one runs out, and more providers give more room.
Some free tiers may train on prompts. They are marked on the Providers panel; use the private profile for anything you would not post publicly.

Project details

License: MIT
Platforms: Linux, macOS, Windows
Source: github.com/dklymentiev/wend-router