Strix Adds Vercel AI Gateway: Setup, Pricing and Security Testing Limits

Strix, the open-source AI penetration-testing tool, added Vercel AI Gateway support in an October 29 announcement. Security teams can configure one Gateway key and switch model routes without opening a separate account for each provider. The practical change is simpler inference management, not a new security model.

Vercel's announcement describes access to more than 400 models and centralized usage tracking. That catalog size is not a claim that every model is equally suitable for a security investigation.

How to configure Strix with AI Gateway

The official Strix setup guide uses two environment variables:

export STRIX_LLM="vercel_ai_gateway/zai/glm-5.3"
export LLM_API_KEY="your-ai-gateway-api-key"

The first part selects Vercel's routing integration. The rest identifies the provider and model in the Gateway catalog. The example uses GLM-5.3, which Strix currently recommends in its guide.

This configures inference; it does not install Strix or authorize a penetration test. Only test systems you own or have explicit permission to assess, with an agreed scope and appropriate controls.

There is a small documentation discrepancy worth keeping in mind. Strix's guide still describes a catalog of more than 100 models, while Vercel's launch page says more than 400. Neither number tells you whether a particular model supports the workflow you need. Check the current model listing and relevant access requirements rather than choosing by catalog size.

What changes from other provider routes

Strix already documents an OpenRouter integration using the same broad pattern: a route in STRIX_LLM and a credential in LLM_API_KEY. Vercel support therefore adds another model-access option rather than introducing multi-provider inference to Strix for the first time.

For a team already managing inference in Vercel, the benefit is keeping security-agent usage alongside other model spending. For a team satisfied with its existing provider setup, the announcement does not establish a reason to migrate on accuracy alone.

A useful comparison would run the same authorized evaluation through each configuration and examine validated findings, runtime and total spend. Changing the route is easy; proving that a different model improves the security workflow takes more work.

Pricing and provider safeguards still apply

Vercel's pricing documentation says token usage has no Gateway markup or platform fee. Charges follow provider list prices. That does not make a Strix run free or give it a fixed price: model choice and tokens consumed still determine inference spending.

The Gateway free tier includes selected models, not its full catalog. Bring-your-own-key access requires purchased credits. Optional Gateway capabilities can also carry separate charges, so zero token markup should not be read as zero possible additional cost.

A shared endpoint does not override model-provider safeguards. Strix's guide explicitly says OpenAI, Anthropic and Gemini models retain their safeguards through the Gateway and directs security-testing users to the providers' respective access programs.

The useful development here is operational consistency. A security team can make model selection less entangled with billing and account setup, while keeping authorization, evaluation and provider requirements as separate decisions. Easier access should make controlled comparisons simpler, not make the controls optional.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders explain developer tools and product launches clearly, with content that makes the technical change understandable and worth sharing.

Book a 30-minute call