Warp's Free Plan Has No AI Included — Here's How to Turn AI On With Your Own API Key (BYOK)

Warp is a terminal with an AI agent built in. The first thing most people hit after installing the free version is: why won’t the AI do anything?

The docs answer it directly: the Free plan does not include bundled AI usage for the Warp Agent. That does not mean free users get no AI — you can bring your own API key. Here is the workable version of what the official documentation says, verified 2026-09-07.

1. What free is missing is the model spend, not the features

Warp’s pricing page lists the Free plan (US$0/month) as including: the modern terminal itself, Warp Agent CLI access, credit reloads at pay-as-you-go rates, bring your own AI inference, limited cloud agents access, individually configured data controls, limited Warp Drive and collaboration access, and limited cloud conversation storage.

Per the docs, using the Warp Agent on Free means one of three routes:

1. Upgrade to a paid plan (Build from US$20/month, including 1,500 credits)
2. Buy add-on credits without subscribing — available on Free, though at a higher rate than paid plans
3. Bring your own inference: BYOK, a custom inference endpoint, or a SuperGrok / X Premium subscription

The payoff for route 3 is stated explicitly: requests routed through your own inference don’t consume Warp credits.

2. What BYOK supports

Per the docs: API keys for OpenAI, Anthropic and Google. xAI’s Grok works differently — you connect a SuperGrok subscription instead of entering a key.

One question the docs answer head-on: can you sign in with a ChatGPT or Claude consumer subscription? No. Those cannot be connected the way SuperGrok can. You need a developer API account, billed by usage rather than a flat monthly fee.

Warp also distinguishes four similar-sounding options:

An eligibility limit worth reading: the docs state BYOK and custom inference endpoints are available to individual users and organisations with 10 or fewer employees, subject to Warp’s Terms of Service. Larger companies need Business or Enterprise. That is a contractual limit, not a technical one.

3. The actual setup

Following the locations named in the docs:

1. Open Warp, go to Settings, and search for API keys.
2. Add your OpenAI / Anthropic / Google key. (For Grok: Settings > Agents > Warp Agent, choose to connect your SuperGrok subscription, and Warp opens your browser to finish.)
3. In the model picker, select the model showing the key icon — that icon means the request routes through your key.
4. To rotate or replace keys later, search API keys in Settings again.

The docs are specific about the billing consequence: when you pick a key-icon model, inference is billed through your provider account rather than your AI credits, and Agent Mode prioritises BYOK over any available Warp credits.

4. What happens when it fails (this is the money part)

The documented behaviour:

• Invalid key: Warp notifies you and halts the request.
• Usage or rate limits hit: Warp will not retry using credits.
• By default, Warp does not fall back to your credits when a BYOK request fails.
• You can opt into Warp credit fallback, which routes a failed BYOK request to a Warp-provided model. Even then, Warp prioritises your keys first and only spends credits when necessary.

In short, the default is the safe one — nothing spends your money quietly. Turn fallback on only if uninterrupted flow matters more to you than a predictable bill.

5. Three things that catch people out

1. Not every AI feature routes through BYOK. Per the docs: Active AI Recommendations do not consume credits (included on Build and higher); Codebase Context does consume Warp credits; Cloud Agents consume credits too, and BYOK keys are stored locally so cloud-hosted runs cannot use them. Setting up BYOK does not make cloud agents free.
2. On Business and Enterprise, local agent runs using BYOK still consume platform credits covering Warp’s platform layer — run lifecycle, integrations, observability. Individual users are not affected by this.
3. Data retention responsibility shifts to you. Warp states it holds Zero Data Retention agreements with its contracted providers, but retention on the provider side depends on your own account settings when you use your own key, and Warp cannot enforce ZDR for those requests. The docs also state plainly: Warp itself never stores your LLM API keys.

6. So is the free plan worth it?

If you already hold an API account — many developers do — Free plus BYOK is a genuinely good combination: the terminal and Agent CLI work, and AI spend goes straight to the model provider by usage instead of an extra US$20 monthly subscription.

If you have no API account and no desire to manage three provider bills, Build’s 1,500 credits are the convenience tax: US$20 for an environment you never have to configure. Both are defensible; the difference is whether you want to manage keys yourself.

Based on direct reads of the Warp BYOK documentation and the Warp pricing page, verified 2026-09-07. Features and eligibility terms can change; the official pages are authoritative. Plan pricing is tracked on our Warp tool page.