AI Gateway vs. Vercel AI Gateway: Managed gateway at the vendor or infrastructure component in your own tenant?

Key facts (as of 26 September 2026):

• Vercel AI Gateway is, per its documentation, an exclusively managed service by Vercel Inc. at ai-gateway.vercel.sh with no documented self-operation, offering regional inference in the zones us and eu; some features are tied to the Pro and Enterprise platform plans.
• The innFactory AI Gateway is deployed into the customer’s cloud tenant and bundles cost centers with per-member caps, budgets in euros and a companion that measures the agent time of Claude Code, Codex and OpenCode per repository, including subscription usage.
• Per the documentation (as of 26 September 2026), budgets in Vercel AI Gateway are a soft cap in US dollars at the scopes team, project, API key and user, BYOK spend does not count toward budgets, and what is measured are requests through the gateway; in the innFactory AI Gateway, per-member caps in a cost center are hard stops in euros, and the aigw companion measures the agent time of Claude Code, Codex and OpenCode including subscription usage.
• Vercel AI Gateway fits when you build with the AI SDK or on Vercel, want to start without your own infrastructure and budgets as soft caps in US dollars are sufficient; the innFactory AI Gateway fits when the gateway should run in your own tenant with no gateway operator in the request path, with hard budget limits per member, roles from Entra ID or Keycloak and agent time per repository.

Vercel describes its product on the product page as “The AI Gateway for developers” and in the documentation as a “managed gateway” that centralizes credentials, request logs, spend budgets, routing and provider failover. By its own account, it first powered v0.app and was declared generally available in 2025.

Vercel AI Gateway is a good product: very broad model and provider coverage, mature provider routing, a setup command for roughly thirty coding agents and native AI SDK integration - without your own infrastructure.

The real decision is an operating question: should all of your company’s AI requests pass through infrastructure operated by Vercel, or should the gateway live in your own cloud tenant - with hard budget limits, roles from your own identity provider and measured agent time of coding agents, even where no gateway sits between tool and model?

What Vercel AI Gateway is

Vercel AI Gateway is operated by Vercel Inc.; its privacy policy states a contact address in Covina, California. The gateway itself is a closed, managed product; the client side is the AI SDK with the @ai-sdk/gateway package. Some features are documented as tied to the Pro and Enterprise plans. Ownership changes are not documented on the pages retrieved.

Operating model: SaaS, self-hosting or infrastructure in your own tenant

Vercel AI Gateway is offered exclusively as a managed service at https://ai-gateway.vercel.sh; a container image or binary for self-operation is not documented on the pages retrieved. The application may run anywhere: “Your application does not need to run on Vercel.” The data path, however, always passes through Vercel’s infrastructure, including with your own provider keys (BYOK).

Vercel documents regional inference with the zones us and eu, set per request; not every model supports both. Without an explicit region: “If you request no region, AI Gateway routes your data globally.” The region setting pins where inference runs, not where the request reaches the gateway. Vercel itself writes that regional routing answers the location question but does not by itself make a workload GDPR-compliant. Zero Data Retention is the documented default.

The innFactory AI Gateway is not hosted by us but deployed into your own cloud tenant: Azure Container Apps or AKS as the reference path, STACKIT for the sovereign route, any Kubernetes. The backend is a Rust binary in a container plus a Next.js dashboard, with PostgreSQL, MongoDB or Azure Cosmos DB. The gateway and its data stay in your tenant; innFactory is not in the request path to the model providers; innFactory GmbH supports the rollout. The gateway has been in production with first customers since August 2026, currently version 1.0.0-rc16.

Identity, SSO and roles

According to the documentation, Vercel AI Gateway authenticates via API keys or Vercel OIDC tokens, generated for a Vercel project and valid for twelve hours. SSO for people is a matter of the Vercel platform: SAML Single Sign-On is documented for Enterprise teams, Pro teams can add it as an add-on; Directory Sync (SCIM) is reserved for Enterprise teams. The role model corresponds to the Vercel team roles Owner, Member, Developer, Security, Viewer and Billing, supplemented by the “AI Gateway Budget Manager” permission.

The innFactory AI Gateway authenticates against Microsoft Entra ID or a generic OIDC provider (Keycloak documented, AWS Cognito claims handled). There are seven roles: Owner, Admin, Auditor, Finance, Security, Member and Guest; guests are created for logins outside the allowed e-mail domains and get no proxy access. IdP groups map to six roles, and models, budgets and rate limits can be set per person - in every license tier.

Cost control and AI FinOps

Vercel documents budgets at exactly four scopes: team, project, API key and user. A budget is, in the vendor’s words, “a soft cap, not a hard limit”: the check runs at the start of each request, the crossing request still completes, then the gateway responds with HTTP 402. Refresh periods are daily, weekly, monthly or none, each in UTC. Budgets are set in US dollars; no other currency is mentioned. BYOK requests do not count toward budgets: “A budget can’t be used to cap BYOK spend.” Alerts at 50, 75 and 100 percent go by e-mail. Every response includes gateway.cost and a generationId; the Logs page exports CSV or JSON.

The innFactory AI Gateway works with independent gates: organization (monthly), cost center (monthly plus daily and weekly windows, pool or per-user mode with per-member caps), user and API key. Every exhausted gate responds with HTTP 402 and E4006; optionally a fallback model takes over, while per-member caps remain hard stops. Add TPM and RPM limits, alerts at 80 and 90 percent and a budget_exceeded event for flows and webhooks. Amounts are displayed in euros at the ECB daily rate, with ai-prices.eu as price basis. Every buffered response carries the x-aigateway-response-cost header; Insights break down usage by user, cost center, model, tool and repository, and the CSV export includes agent and LLM time.

Coding agents, subscriptions and agent time

This is where Vercel is strong. The documentation lists roughly thirty coding agents, including Aider, Claude Code, Cline, Codex, Cursor, OpenCode and Zed. The command npx vercel ai-gateway setup detects installed agents and writes their configuration files; Claude Code, Codex and Cursor receive dedicated compatibility endpoints. Measured is what passes through the gateway: requests, tokens and costs per key, project or user. According to Vercel’s documentation (as of 26 September 2026), a local companion that measures agent time or captures usage via subscriptions such as Claude Max or a Codex plan is not described; nor is MDM distribution of a client component.

The innFactory AI Gateway provides proxy setups for six tools: Claude Code, Codex, OpenCode, Cowork, VS Code Continue and VS Code Copilot (BYOK). Each developer clones an admin template into an own setup with own key and cost center; aigw run claude|codex|opencode launches the tool with a short-lived proxy token. The aigw companion runs as a background service on Linux, macOS and Windows. It reads the local session logs of Claude Code, Codex and OpenCode and reports agent time, LLM time and token classes per repository, user, tool and cost center. It detects whether a session ran through the gateway or on the tool’s subscription; subscription usage is valued at list price, shown separately, never billed or budgeted. Rollout uses device-code login or enrollment tokens through Intune, a macOS policy or a Linux script; only a code-frozen allow-list without prompts is transmitted. Cursor, Windsurf, Aider and Cline are currently not covered by the companion.

Providers, dialects and models

On breadth, Vercel is ahead. The documentation’s provider table names several dozen providers, from Anthropic, OpenAI, Google, Azure and Bedrock to specialists such as Groq or Cerebras; the vendor speaks of “hundreds of models”. From Europe, Mistral appears in the table; STACKIT, IONOS and Aleph Alpha are not listed there as providers. Inbound, the OpenAI Chat Completions API, the Responses API and the Anthropic Messages API are implemented; as modalities Vercel names text, image and video generation, speech, realtime, embeddings and reranking, among others.

The innFactory AI Gateway supports 13 provider types, including OpenAI, Azure OpenAI, Azure AI Foundry, Anthropic (direct or Vertex), Gemini (including the Vertex EU endpoint), Mistral, AWS Bedrock, STACKIT, IONOS, Ollama, ElevenLabs and OpenAI-compatible endpoints. Inbound: three dialects (OpenAI, Anthropic, Gemini), realtime WebSockets for OpenAI, Azure OpenAI, ElevenLabs and Gemini Live, plus audio, embeddings, images and OCR. If you need many niche providers, Vercel has the larger catalog; if you want STACKIT or IONOS, you will find them with us.

Governance: guardrails, MCP, routing and Flow Builder

As security features, Vercel documents Zero Data Retention (on by default), a control against prompt training, a model allowlist and a team-wide provider allowlist for the Pro and Enterprise plans. Routing is well developed: by default the gateway selects providers by uptime and latency; per request, order, only and sort by cost, ttft or tps can be set, plus model fallbacks. Guardrails such as PII redaction or prompt-injection detection, shadow mode, a separate audit log, generic webhooks and a visual flow editor are not documented on the pages retrieved. Vercel MCP manages Vercel’s own platform; a gateway-side MCP proxy for your applications’ tools was not described on the AI Gateway pages retrieved.

The innFactory AI Gateway ships nine guardrail rule types, from secret detection and PII redaction via prompt injection to an external webhook, with shadow mode and a GDPR preset. The MCP proxy exposes Streamable HTTP servers with eight upstream auth modes (among them per-user OAuth PKCE, Entra on-behalf-of, AWS SigV4), access policies, toolsets and guardrails on tool arguments. Routing includes an auto-router by prompt complexity, fallbacks and deployment groups. The Flow Builder is a visual graph with nine strategies, guardrail nodes with pass/block/fail branches and event triggers for webhooks; a flow is callable like a model and bills the same budgets. Configuration is YAML, the Terraform provider or natural language through a built-in MCP server.

What Vercel AI Gateway does better

• Coding-agent setup: one command configures roughly thirty agents, with dedicated endpoints for Claude Code, Codex and Cursor.
• Catalog breadth: several dozen providers and, according to the vendor, hundreds of models, including image, video and reranking models.
• AI SDK integration: according to the documentation, the default provider of the AI SDK as soon as a model is specified as a string.
• Provider routing and operations: dynamic selection by uptime and latency, sorting by cost, time-to-first-token or throughput - without your own proxy, database or routing control plane.
• Billing and certifications: billing on the existing Vercel invoice; platform-wide, Vercel names SOC 2 Type 2 and ISO 27001:2022.

When Vercel AI Gateway is the right choice

• You build with the AI SDK or on Vercel, want to start without your own infrastructure and need a very broad model catalog.
• Your data-protection concept allows a third-party data path, and you set the inference region deliberately.
• Budgets as soft caps in US dollars are sufficient for your control needs.
• Your coding agents run through gateway keys anyway rather than through subscriptions.

When the innFactory AI Gateway is the right choice

• The gateway is to run in your own tenant (Azure, STACKIT, Kubernetes), with no gateway operator in the request path.
• Roles and cost centers come from Entra ID or Keycloak, in every license tier.
• You need hard budget limits per member, euro display and a fallback model instead of an error message.
• You want to see the agent time of Claude Code, Codex and OpenCode per repository and cost center, including subscription usage.
• Guardrails, MCP proxy and Flow Builder are to be one stack with CompanyGPT; STACKIT or IONOS are to be connected as providers.
• The license is a monthly flat rate by user count (199 to 749 EUR excluding VAT); token costs are billed by the providers directly.

Comparison at a glance

Conclusion

Vercel AI Gateway solves a clear problem well: development teams get a very broad model catalog, robust provider routing and a setup for the common coding agents, without their own infrastructure. Whoever accepts a data path through a managed service has a coherent solution.

The innFactory AI Gateway answers a different question: how does a company get AI governance and FinOps into its own tenant, with roles from its own identity provider, hard budget limits and a measurement that also makes agent time via subscriptions visible? Both answers are legitimate; which fits is decided by operating model, data path and the degree of control you need.

Natively integrated with CompanyGPT, one reusable stack

The innFactory AI Gateway is natively integrated with CompanyGPT: the same budgets, cost centers and guardrails apply to chat, agents and add-ons. Identity via Entra ID or Keycloak, cost centers, guardrails and the MCP registry are one stack, reused across CompanyGPT, the gateway and your own applications. The gateway can also run standalone. Further comparisons are available under all comparisons, the product overview under AI Gateway.

Sources

Retrieved on 26 September 2026:

• Vercel AI Gateway product page — https://vercel.com/ai-gateway
• Vercel AI Gateway documentation, overview — https://vercel.com/docs/ai-gateway
• Vercel blog: AI Gateway is now generally available — https://vercel.com/blog/ai-gateway-is-now-generally-available
• Vercel Privacy Policy — https://vercel.com/legal/privacy-policy
• Regional Inference — https://vercel.com/docs/ai-gateway/security-and-compliance/regional-inference
• Security and Compliance — https://vercel.com/docs/ai-gateway/security-and-compliance
• Budgets — https://vercel.com/docs/ai-gateway/observability-and-spend/budgets
• Bring Your Own Key — https://vercel.com/docs/ai-gateway/authentication-and-byok/byok
• OIDC Authentication — https://vercel.com/docs/ai-gateway/authentication-and-byok/oidc
• Logs — https://vercel.com/docs/ai-gateway/observability-and-spend/logs
• Pricing — https://vercel.com/docs/ai-gateway/pricing
• Coding Agents — https://vercel.com/docs/ai-gateway/coding-agents
• Anthropic Messages API — https://vercel.com/docs/ai-gateway/sdks-and-apis/anthropic-messages-api
• OpenAI Chat Completions — https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-chat-completions
• Models and Providers — https://vercel.com/docs/ai-gateway/models-and-providers
• Provider Options — https://vercel.com/docs/ai-gateway/models-and-providers/provider-options
• Provider Filtering and Ordering — https://vercel.com/docs/ai-gateway/models-and-providers/provider-filtering-and-ordering
• SAML Single Sign-On — https://vercel.com/docs/saml
• Directory Sync — https://vercel.com/docs/directory-sync
• Vercel Security and Compliance — https://vercel.com/docs/security/compliance

Note on the information: All statements about other vendors’ products are based on their publicly available documentation as of the stated date. Vendors continuously develop their products, features and terms — the vendor’s current information always prevails. If any statement appears outdated or incorrect to you, please drop us a line at info@innfactory.ai; we will review and correct it promptly. This comparison does not replace legal or data-protection advice in individual cases.

Further reading

• AI Gateway – product overview, proxy and companion mode
• CompanyGPT – the platform the gateway is natively integrated with
• Claude Code as an AI harness – building blocks, models and enterprise options
• AI harness explained – fundamentals of AI agents and harnesses
• All AI Gateway comparisons – LiteLLM, OpenRouter, Portkey, Cloudflare, Kong, Bifrost, TrueFoundry, Helicone, Vercel
• AI Gateway vs. OpenRouter
• AI Gateway vs. Portkey (PRISMA AIRS)
• AI Gateway vs. Cloudflare AI Gateway