Claude Code Free: Run It and AI Agents for $0

Published November 10, 2026.

Claude Code is Anthropic's terminal coding agent, and as of November 10, 2026, the official Free Claude plan does not include it. If you want to run Claude Code free without paying subscription or API token fees, you can point the tool at free third-party models using Wend router. A second path is Flint Agent, an open-source terminal assistant that connects directly to the same free endpoints. Here is how each route works and what it costs.

Can I use Claude Code for free through official Anthropic plans?

The official Claude pricing page at claude.com/pricing lists four tiers: Free, Pro, Max, and Ultra. The Free tier costs $0, but Anthropic does not include Claude Code in that plan. Accessing the coding agent directly through Anthropic always requires a financial commitment, either through a recurring subscription or direct API billing.

The Pro plan costs $17 a month when billed annually, which amounts to $200 billed up front, or $20 a month when billed monthly. Claude Code is included in the Pro tier. The Max plan costs $100 a month and includes Claude Code. The Ultra plan costs $200 a month and also includes Claude Code. On each of these paid tiers, Claude Code shares usage limits with the rest of Claude. If you prefer not to commit to a monthly subscription, you can run Claude Code pay-as-you-go at API rates through an Anthropic Console account. With Anthropic's own models, Claude Code is never free. For a detailed breakdown of plans and token pricing, consult the guide at klymentiev.com/blog/claude-code-pricing on our site.

How can you set up Claude Code free with Wend router?

If you want to know how to use Claude Code for free, the operational mechanism relies on changing the target endpoint. Claude Code is built to read three environment variables from your shell environment: ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, and ANTHROPIC_MODEL. When you point ANTHROPIC_BASE_URL to an address that speaks the Anthropic API and serves as a router in front of free models, Claude Code runs on those free models. Under this configuration, your Anthropic bill is $0.

Wend router is a free, open-source LLM router published under the MIT license, hosted at github.com/dklymentiev/wend-router on GitHub. It presents one local address compatible with both the OpenAI API and the Anthropic API in front of many LLM providers. When a request arrives, the router selects a model that is up and fits the job. If a provider fails, runs out of quota, or stalls, the router automatically moves the request to another model on its own. This failover behavior ensures that the terminal agent does not halt during execution.

Pointing the agent at a local router changes the models that answer. Claude Code sends the prompt, but the free models behind the router supply the answers.

Wend router ships as a single program for Linux, macOS, and Windows, and includes a built-in dashboard. The system prioritizes free tiers first. Paid models are used only when you explicitly allow them within an established budget. Some providers work without any API key, meaning the router answers requests immediately before you add a single key. Adding more providers grants access to more models and creates more room before reaching daily limits. You can add a free key for each provider on the dashboard's Providers panel. A directory of free LLM APIs and instructions on where to get their keys is available at klymentiev.com/blog/free-llm-api on our site. All provider keys are stored encrypted on your own machine.

Installation requires a single command without root or administrator rights. On Linux or macOS, run the installation script directly:

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

On Windows PowerShell, run the installer with:

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

Everything is stored inside a single directory, located at ~/wend on Linux and macOS, or %LOCALAPPDATA%\wend on Windows. The script launches the router, sends one test request, and opens the dashboard at 127.0.0.1:8888/dashboard in your browser. The router can run on your laptop or on your own server. By default, it listens exclusively on 127.0.0.1. If you need to serve other machines across your network, set ROUTER_ADDR=0.0.0.0:8888 together with ROUTER_TOKEN=<secret>.

When sending requests from an agent, the model parameter can reference a router profile name. The router then restricts its selection to the models assigned to that profile and switches between them as needed. The profile named tools selects models that call tools well, and it is recommended for coding agents such as Claude Code. The no-think profile selects models without a reasoning phase, choosing the quickest first to deliver simple, fast answers. The vision profile directs queries to models that read images. The auto profile includes every model, ordering by strongest first, whereas fast prioritizes the quickest to start. The private profile restricts routing to providers that do not train on prompts. You can also build a custom profile on the dashboard under New profile, such as a profile named think that requires reasoning models for harder questions.

To connect Claude Code on Linux or macOS, export the following environment variables in your terminal before starting the agent:

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

On Windows PowerShell, set the environment variables using PowerShell commands:

$env:ANTHROPIC_BASE_URL = "http://localhost:8888"
$env:ANTHROPIC_AUTH_TOKEN = "wend"
$env:ANTHROPIC_MODEL = "tools"
claude

The authentication token can be set to any arbitrary string because the router does not validate tokens unless you have configured ROUTER_TOKEN.

In a test performed on November 10, 2026, using a clean install of Wend router 0.2.0 without any provider keys added, 17 free models were up immediately. Claude Code was started with ANTHROPIC_MODEL=tools and asked to create a file. It completed the file creation task in 19 seconds at a cost of $0. In the same test session, the think profile answered a classic trick question correctly, determining that when a bat and a ball cost 1.10 and the bat costs 1.00 more than the ball, the ball costs 0.05. The vision profile described a picture of a red square on white correctly. Open WebUI pointed at the router listed the router profiles as models and answered an image question through vision.

Running free Claude Code through an external router involves specific operational limits. The models providing answers are not Anthropic's Claude models, but the third-party free models routed behind the local address. These models are weaker than Claude's best models for hard work. In addition, free tiers have daily limits per provider. While the router shifts requests to another model when a provider runs out of quota, heavy usage can exhaust available tiers. Furthermore, certain free tiers may train on prompts; these providers are marked on the Providers panel, and you should use the private profile for anything you would not post publicly.

Can you use Flint Agent as a free alternative?

If your goal is to use an automated coding assistant in the terminal without paying subscription costs, you do not necessarily need Claude Code. Flint Agent is an open-source AI agent developed for the terminal, accessible at flintagent.dev, and executed using the flint command. Its source is on GitHub under the MIT license: github.com/dklymentiev/flint-agent.

Flint Agent works with Wend router out of the box, and the router dashboard features a dedicated Flint tab within its Connect section. To configure Flint Agent, point its OpenAI compatible interface to the local router:

export FLINT_PROVIDER=openai
export FLINT_BASE_URL=http://localhost:8888/v1
export OPENAI_API_KEY=wend
flint --model tools

The last line starts the agent on the tools profile. In a test conducted on November 10, 2026, against the same keyless Wend router installation, Flint Agent completed its self-check routine and finished a small file task in 6 seconds at a cost of $0. Flint Agent fits users who do not require Claude Code itself, providing an agent that operates on free models without a subscription.

More about Flint Agent on its project page: klymentiev.com/projects/flint.

Any other agent or chat application supporting an OpenAI endpoint can connect to the router using the same method. You set the base URL to http://localhost:8888/v1, pass any text as the API key, and specify a profile name as the model. Each tool can operate on an independent profile, such as a terminal coding agent using tools, an interface like Open WebUI using vision to inspect images, and a fast background utility using no-think.

Which route fits your workflow?

Choosing an approach depends on whether you require official Claude models, whether you need Claude Code's terminal interface, or whether an open-source agent satisfies your project requirements.

When evaluating these paths, consider model performance constraints. For hard, long coding work, a paid plan using Anthropic's models is still the stronger tool. When working with free tiers through a router, adding more provider keys expands your daily limits and gives the routing engine more model options during automated failover.

Going to production. Free tiers fit learning, side projects and prototypes. When real users, deadlines or customer data depend on the answers, use paid models: they are stronger, and their limits are ones you pay to raise rather than ones that run out mid-task. Wend router can stay in front of them: add a paid provider's key and allow paid models in a profile, within a daily budget you set.

Frequently asked questions

Can I run Claude Code with zero API keys added?

Yes. Some providers in Wend router work without any API key. On a clean installation of Wend router 0.2.0 tested on November 10, 2026, 17 free models were available immediately without adding keys. In that test, Claude Code completed a file creation task in 19 seconds at a cost of $0. Adding keys from free provider tiers on the Providers panel is optional and increases your available daily quota.

Is Wend router free?

Yes. Wend router is free and open source under the MIT license. It uses free tiers first, and paid models only when you allow them, within a budget you set.

Can Wend router run on my own server instead of my laptop?

Yes. It can run on your laptop or on your own server. By default it listens only on the machine itself (127.0.0.1). To serve other machines, set ROUTER_ADDR=0.0.0.0:8888 together with ROUTER_TOKEN=<secret>.

What are the limits of running Claude Code through free tiers?

Free models behind a router are weaker than Claude's best models for hard work. In addition, free tiers have daily request limits per provider. When a provider runs out of quota, Wend router shifts requests to another available model on its own, but persistent usage can consume multiple free tiers. Finally, some free providers may train on prompts, which the router marks on the Providers panel. The private profile should be selected for any data you would not publish openly.