Configure model routing, failover, and cost tracking through the Vercel AI Gateway unified API.
npx skills add https://github.com/AarnavBaddam/skills --skill ai-gateway-aarnavbaddam
Please help me install this Agent Skill. Skill: ai-gateway Source: https://github.com/AarnavBaddam/skills/tree/main/ai-gateway Command: npx skills add https://github.com/AarnavBaddam/skills --skill ai-gateway-aarnavbaddam
provider/model slugs through one endpoint. - Failover and Provider Ordering: Configure order, only, and fallback models so requests automatically reroute when a provider is unavailable. - Cost Controls and Observability: Apply per-user rate limits, tag-based budget attribution, cache-control headers, and audit logging for production workloads. - Use Case: You are building a multi-tenant SaaS chat feature. Use this Skill to set up OIDC authentication via vercel env pull, route anthropic/claude-sonnet-4.6 with an OpenAI fallback, tag requests per feature, and enforce per-user rate limits. ## Quick Start Ask the AI to set up Vercel AI Gateway in your project with OIDC authentication and a generateText call using the openai/gpt-5.4 model slug.High-intent search queries and answers about installing and using this skill.
Pass a provider/model string like openai/gpt-5.4 to the model parameter in generateText, and the AI SDK routes it through the gateway automatically. Use the gateway() wrapper only when you need providerOptions.gateway for routing, failover, or tags.
Run vercel link, enable AI Gateway in your project settings, then run vercel env pull to provision a VERCEL_OIDC_TOKEN in .env.local. Alternatively, set AI_GATEWAY_API_KEY as a static key for CI or non-Vercel environments.
Use the gateway when you need failover, cost tracking, per-user rate limiting, or audit logging across providers. Use a direct provider SDK for provider-specific features like Anthropic computer use, self-hosted models, or custom HTTP transport control.
Yes, configure providerOptions.gateway with order to prioritize providers, only to restrict providers, and models to specify fallback models. When a provider errors or exceeds quota, requests automatically reroute to the next option.
HTTP 429 means a user exceeded configured rate limits; check the retry-after response header. HTTP 402 means a hard budget limit was reached; handle it by degrading gracefully with a fallback response.
No, tokens are billed at exact provider list price with zero markup, whether using Vercel-managed keys or Bring Your Own Key. Every team gets $5 of free credits monthly, with pay-as-you-go top-ups beyond that.
vinhnguyenn020325-cmdManage Vercel environment variables with .env files and the vercel env CLI.
Vivekmano27Integrates LLM APIs into production services with prompt templates and routing.
foolishimpGenerate per-REQ dashboards and alerts for Datadog, Prometheus, and Splunk.
ax-llmWrite Go provider clients and model-routing code for AxAI.