Warp Lets 800K Developers Run Any OpenRouter Model Inside Their Terminal
Warp's agent now accepts any OpenRouter-compatible URL, giving developers instant access to 300+ models with a single alias command
- Warp's agent now accepts OpenRouter as a custom inference endpoint, unlocking 300+ models from a single URL config.
- Setup takes three fields: your OpenRouter base URL, a model slug, and an alias for quick
/modelswitching in-terminal. - No Warp credits consumed when using BYOK or custom endpoints — all costs bill directly to your OpenRouter account.
- API keys are stored locally only and never synced to Warp's cloud servers.
- Available on all plans including Free for individuals and teams of 10 or fewer, with no minimum spend.
- Part of a broader open-inference push: Warp also plans local model support via a Rust harness and Agent Client Protocol integration.
Warp, the agentic terminal trusted by over 800,000 developers, just made its built-in AI agent dramatically more flexible. You can now point Warp's agent at your own OpenRouter URL, unlocking a catalog of hundreds of models from Anthropic, OpenAI, Google, DeepSeek, Kimi, and more, all without leaving the terminal. The change is live now and available on every plan, including Free.
The problem it solves
Until now, Warp's agent ran on Warp's own bundled inference, meaning you were limited to the models Warp chose to support. That was fine for most users, but it left a real gap: developers who already had OpenRouter credits, wanted to experiment with a newly released model, or needed to route through a specific provider for cost or compliance reasons had no path forward inside Warp.
The model layer needed to be more flexible. Developers should be able to use Warp Agent with the inference setup that works best for them, including providers or models Warp does not support natively yet, specific model configurations, custom routing behavior, and existing credits with model providers.
What OpenRouter actually is
If you haven't used it before: OpenRouter is a unified API gateway that provides access to 315+ AI models from every major provider through a single, OpenAI-compatible API. Instead of managing separate API keys and billing for each provider, you use one API key and one credit balance. Think of it as a universal adapter for LLM inference.
The reliability angle matters too. OpenRouter routes requests across multiple providers. If one provider is unavailable or rate-limited, OpenRouter can fail over to another, keeping your coding sessions uninterrupted. For agentic workloads that can run for minutes at a time, that kind of resilience is genuinely useful.
How to set it up
The setup is minimal.
Enable this by adding your OpenRouter API key in settings under the custom inference endpoint section. Once configured, you can switch models using the /model command.
You also assign a short alias to each model configuration, so flipping between, say, a cheap DeepSeek run and a Claude Opus session is a single slash command away.
Here's the mental model for the three configuration fields Warp asks for:
- URL , your OpenRouter base endpoint (
https://openrouter.ai/api/v1) - Model , any OpenRouter model slug, e.g.
anthropic/claude-opus-4ordeepseek/deepseek-r1 - Alias , a short name you type after
/modelto switch to that config instantly
OpenRouter is just one option. Warp supports any custom inference endpoint that follows the OpenAI Chat Completions API standard. This includes popular aggregators and gateways like OpenRouter, LiteLLM, and z.ai, as well as specific model providers like DeepSeek.
What it costs
When you use your own API key through a custom inference endpoint like OpenRouter, Warp does not consume your AI credits. Instead, all inference costs are billed directly through your OpenRouter account. This provides full control over your billing and allows you to bypass Warp's internal credit system.
On the OpenRouter side, OpenRouter charges per-token, passing through model provider pricing. You add credits to your account and pay only for what you use. Input tokens and output tokens are priced separately. There are no monthly fees or minimum commitments. OpenRouter also offers dozens of free models with zero per-token cost, though free models have rate limits of 20 requests per minute and 200 requests per day.
For individual developers and small teams, BYOK and custom inference endpoint usage are free in Warp. Whether you are on Free, Build, or Max, Warp will not charge Warp credits when you bring your own API key or connect your own compatible inference endpoint, as long as the usage is for an individual or a company with 10 people or fewer.
The security story
A reasonable concern when pasting API keys into a terminal app: Warp stores your API keys locally on your device. These keys are never synced to Warp's cloud servers or retained by the company. Because the keys remain local, they are not available for cloud-hosted agent runs, which will continue to consume Warp credits rather than your personal keys. That last point is worth noting if you rely on Warp's cloud agents for background tasks.
Where this fits in the bigger picture
This OpenRouter hook is part of a broader push by Warp toward what they call an open agentic development environment. Warp recently open-sourced the Warp client because they want Warp to become more open, customizable, and shaped by how developers actually work. Bringing your own inference to power the Warp Agent is a part of that.
The roadmap signals even more flexibility ahead. Warp is planning to add a lightweight Rust client harness to their open-source repo, so Warp can connect directly to local models without routing through Warp's servers. They also plan to support Agent Client Protocol, so developers can bring other agent harnesses into Warp's terminal UI.
The practical upshot: if you're already paying for OpenRouter credits, or if you want to run a model that Warp doesn't natively support yet, there's no longer any reason to leave the terminal to do it. Set your URL, name your alias, and type /model.