Can I self-host Cloudflare AI Gateway?

According to the documentation (as of 26 September 2026), Cloudflare AI Gateway is a service operated by Cloudflare; a Docker image, Helm chart or any other form of self-hosting is not documented. The innFactory AI Gateway is deployed as a container into your own cloud tenant - Azure Container Apps, AKS, STACKIT or any Kubernetes.

Does Cloudflare AI Gateway offer EU data processing?

In the compatibility table of the Data Localization Suite, Cloudflare marks AI Gateway as not compatible with Regional Services and Geo Key Manager. We therefore did not find an EU-specific processing option for the product. With the innFactory AI Gateway, gateway, dashboard and data run in the customer tenant, because the gateway runs there.

How do the budgets differ?

According to its documentation, Cloudflare sets spend limits per model, provider or metadata dimension with a hard stop via HTTP 429 or a fallback through Dynamic Routing. The innFactory AI Gateway has cost centers as objects of their own with pool or per-user mode, per-member caps and independent gates on organization, cost center, user and API key.

Does Cloudflare AI Gateway measure the working time of coding agents?

Cloudflare documents integration guides for Claude Code, Claude Desktop, GitHub Copilot CLI, OpenAI Codex, OpenCode and Pi. What is captured are the requests that pass through the gateway. A local companion measuring agent time or subscription usage is not described in the documentation (as of 26 September 2026); the innFactory AI Gateway ships the aigw companion for this.

Does Cloudflare AI Gateway have a visual flow builder?

Yes. According to the documentation, Dynamic Routing offers a visual editor with the nodes Start, Conditional, Percentage, Model, Rate Limit, Budget Limit and End, plus versioned routes with deploy and rollback. The Flow Builder of the innFactory AI Gateway adds guardrail nodes with pass/block/fail branching and event triggers that fire webhooks.

Cloudflare AI Gateway Alternative: AI Gateway Compared

Key facts (as of 26 September 2026):
Cloudflare AI Gateway is, per its documentation, a closed-source managed edge service operated by Cloudflare with no documented self-hosting, available in the plan tiers Workers Free, Workers Paid and Enterprise.
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.
In the compatibility table of the Data Localization Suite, Cloudflare marks AI Gateway as not compatible with Regional Services and Geo Key Manager, and per the documentation (as of 26 September 2026) what is captured is what passes through the gateway; with the innFactory AI Gateway gateway, dashboard and data run in the customer tenant and innFactory is not in the request path, and the aigw companion additionally measures agent time and subscription usage.
Cloudflare AI Gateway fits when your application already runs on Cloudflare Workers, you do not want to operate a gateway and processing through Cloudflare’s global network fits your data-protection assessment; the innFactory AI Gateway fits when the gateway should run in your own cloud tenant with roles from Entra ID or Keycloak, cost centers with per-user budgets in euros and the agent time of coding agents.

Cloudflare describes its AI Gateway in the documentation as “Observe and control your AI applications” and on the product page as “An intelligent control plane for your AI applications”. Per the vendor, a Cloudflare account and one line of code suffice.

Cloudflare AI Gateway is a strong product, especially for teams already on Cloudflare: nothing to operate, a visual Dynamic Routing editor with versioning, spend limits with a hard stop.

The real decision is an architecture question: AI requests through a Cloudflare-operated edge service — or through an infrastructure component in your own cloud tenant, connected to your identity system, that also captures the work of coding agents outside the proxy? Comparison as of 26 September 2026.

What Cloudflare AI Gateway is

Cloudflare AI Gateway is a closed-source service operated by Cloudflare; per the terms of service, the vendor is Cloudflare, Inc., San Francisco. An open-source release is not documented; per its description, the cloudflare/ai repository contains “Packages and examples for building AI-powered applications on Cloudflare” — SDKs and examples, not the service.

The documentation names the plan tiers Workers Free, Workers Paid and Enterprise; per the pricing page, caching and DLP scanning are included on all plans.

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

Cloudflare AI Gateway is a managed edge service: every request passes through Cloudflare’s infrastructure to the model provider. A Docker image or Helm chart for self-hosting is not documented.

Important for European companies: Cloudflare’s Data Localization Suite compatibility table marks AI Gateway as not compatible with Regional Services and Geo Key Manager; we found no EU-specific processing option for this product. Cloudflare’s Trust Hub lists SOC 2 Type II and ISO 27001 for the entire platform, with no AI-Gateway-specific statement found.

The innFactory AI Gateway is deployed as a cloud-native infrastructure component into the customer’s cloud tenant — Azure Container Apps or AKS, STACKIT or any Kubernetes — and operated there as part of the customer’s AI stack, not hosted by innFactory as SaaS. Written in Rust, the gateway runs as a container with PostgreSQL, MongoDB or Azure Cosmos DB; the data path stays in the tenant.

Identity, SSO and roles

Per the Cloudflare One documentation, SSO and SCIM exist at Cloudflare account level — from Okta, Microsoft Entra or Authentik — an account feature, not an AI Gateway feature. At gateway level there are three API token permissions — AI Gateway Read, Run and Edit — which per the documentation cannot be restricted to a single gateway. Users are attributed via request metadata such as user_id. A per-user role model is not documented for the AI Gateway.

The innFactory AI Gateway connects to Microsoft Entra ID or a generic OIDC provider such as Keycloak. Seven roles — Owner, Admin, Auditor, Finance, Security, Member, Guest —, IdP group mapping and cost-center administrators are included in every license tier. Allowed models, budgets and rate limits can be set per person.

Cost control and AI FinOps

Per the documentation, Cloudflare’s spend limits are set by model, provider, custom metadata such as user_id and the reserved field cf.user_id — as an independent budget per value or as a filter, in fixed or rolling daily, weekly or monthly windows. At the limit, the gateway blocks with HTTP 429; with Dynamic Routing it can fall back to another model instead. Per Cloudflare, spend limits are “eventually consistent”, a maximum of 20 rules per gateway applies, and limits are set in dollars. A CSV export and a cost-center concept are not documented.

The innFactory AI Gateway runs four independent gates: organization, cost center (monthly plus daily and weekly windows), user and API key. Cost centers are objects of their own, in pool or per-user mode with hard per-member caps. An exhausted gate answers with HTTP 402 and error code E4006, or with a fallback model. Alerts fire at 80 and 90 percent, and a budget_exceeded event can trigger a webhook. Costs are attributed by user, key, team, cost center, model, provider, tool, repository and billing source — internally in US dollars, displayed in euros at the ECB daily rate. Each buffered response carries the x-aigateway-response-cost header; the CSV export includes agent and LLM time.

Coding agents, subscriptions and agent time

Cloudflare documents guides for Claude Code, Claude Desktop, GitHub Copilot CLI, OpenAI Codex, OpenCode and Pi. For Claude Code, ANTHROPIC_BASE_URL points at the gateway endpoint and the token is passed via ANTHROPIC_CUSTOM_HEADERS. Captured is what passes through the gateway. Per Cloudflare’s documentation (as of 26 September 2026), a local companion measuring agent time or tool-subscription usage is not described.

The innFactory AI Gateway has two building blocks for this. First, proxy setups from admin templates, each developer with own key and cost center — for Claude Code, Codex, OpenCode, Cowork, VS Code Continue and VS Code Copilot. Second, the aigw companion: a background service on the developer machine that reads the 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 — Claude Max, ChatGPT or Codex plan; subscription usage is valued at list price, shown separately and never billed. The upload is a code-frozen allow-list without prompts. Rollout works via Intune, a macOS policy or a Linux script; the binaries are not code-signed yet, and there is no self-update.

Providers, dialects and models

In provider breadth, Cloudflare is ahead. The OpenAI-compatible endpoint documentation names “14+ providers”, including Anthropic, OpenAI, Google, xAI, Groq, Mistral, Cohere, Perplexity, Workers AI, DeepSeek and Cerebras; STACKIT and IONOS are not documented. Inbound dialects are the REST API, the /compat/chat/completions endpoint (required for Dynamic Routing) and provider-native endpoints such as /anthropic; a Gemini dialect and WebSocket realtime were not documented.

The innFactory AI Gateway has 13 provider types, including OpenAI, Azure OpenAI, Anthropic (direct or Vertex), Gemini (AI Studio or Vertex with EU endpoint), Mistral, AWS Bedrock, STACKIT, IONOS, Ollama and ElevenLabs. Inbound it speaks the OpenAI, Anthropic and Gemini dialects, plus realtime WebSockets, audio, embeddings and images.

Governance: guardrails, MCP, routing and Flow Builder

Per the documentation, Cloudflare Guardrails inspect prompts and responses with a policy that flags or blocks, plus an audit trail. Separately, the DLP feature uses the detection engines of Cloudflare’s Data Loss Prevention product on requests, responses and tool-call arguments, with the actions Flag and Block. An MCP proxy is not documented as an AI Gateway feature.

Dynamic Routing is Cloudflare’s strength here: routes are built in the visual editor or as JSON with the seven node types Start, Conditional, Percentage, Model, Rate Limit, Budget Limit and End — for A/B tests, gradual rollouts and quotas per user, project or team. Routes are versioned, with deploy and rollback.

The innFactory AI Gateway applies nine guardrail rule types — including secret detection, PII redaction, prompt injection and external webhook — with shadow mode, mid-stream redaction and a GDPR preset. The MCP proxy exposes Streamable HTTP servers with eight upstream auth modes, including per-user OAuth PKCE and Entra on-behalf-of. The Flow Builder is a visual graph with nine strategies; the difference from Cloudflare is in the node types: guardrail nodes branch into pass, block or fail, and event triggers such as budget_exceeded fire webhooks.

What Cloudflare AI Gateway does better

Nothing to operate. Part of Cloudflare’s global network.
BYOK via the Secrets Store, with provider keys in the Cloudflare dashboard per the documentation.
DLP engines from the Zero Trust product line, included on all plans per the pricing page.
Six coding-agent guides, including Claude Desktop, Copilot CLI and Pi.
Platform-wide certifications. SOC 2 Type II and ISO 27001.
Dynamic Routing with versions. Visual editor, A/B tests, instant rollback.
Provider breadth, e.g. xAI, Groq, Cohere, DeepSeek, Cerebras, Workers AI.

When Cloudflare AI Gateway is the right choice

Your application already runs on Cloudflare Workers, and you do not want to operate a gateway.
Processing through Cloudflare’s global network fits your data-protection assessment.
You want Groq, Cerebras or Workers AI via the same endpoint.

When the innFactory AI Gateway is the right choice

The gateway should run in your own cloud tenant — Azure, STACKIT or Kubernetes — with the data path staying there.
You need roles from Entra ID or Keycloak in the gateway itself, with Auditor, Finance and Security views.
Cost centers with pool or per-user budgets, per-member caps and display in euros are required.
You want the agent time of Claude Code, Codex and OpenCode per repository and cost center — subscriptions included.
Guardrails with shadow mode, webhook nodes, an MCP proxy or providers such as STACKIT and IONOS are required.

Comparison at a glance

Criterion	Cloudflare AI Gateway (per documentation, as of 26 Sept 2026)	innFactory AI Gateway
Operations	Edge service; self-hosting not documented	Container in your own tenant
EU processing	Not compatible with Regional Services/Geo Key Manager	Data path stays in the customer tenant
Identity	SCIM at account level; token permissions Read/Run/Edit	Entra ID/OIDC, seven roles, in every license
Budgets	Spend limits per model, provider, metadata; 429 or routing fallback	Gates on organization, cost center, user, key; per-member caps
Currency	Limits in dollars	USD internally, displayed in euros at ECB rate
Coding agents	Six guides; captures proxied requests	Six proxy setups plus companion for agent time
Guardrails	Flag/block with audit trail; DLP engines	Nine rule types, shadow mode, mid-stream redaction
MCP	Not documented as an AI Gateway feature	MCP proxy with eight auth modes and toolsets
Routing editor	Dynamic Routing, seven node types, versions, rollback	Flow Builder, nine strategies, guardrail and webhook nodes

Conclusion

Cloudflare AI Gateway is a good choice when the application already runs on Cloudflare, nobody wants to operate a gateway and processing through the global network fits the data-protection concept. Routing editor and spend limits are cleanly documented; the provider breadth is larger than ours.

The innFactory AI Gateway is the choice when the gateway should sit in your own tenant, roles and cost centers come from the identity system and the working time of coding agents should be visible too. Both solve the proxy problem; they differ in where control sits.

Natively integrated with CompanyGPT, one reusable stack

The innFactory AI Gateway is natively connected to 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 MCP registry are one stack reused across CompanyGPT, the gateway and your own tools. The gateway also runs standalone. More comparisons: all comparisons; product page: AI Gateway.

Sources

Retrieved on 26 September 2026:

Documentation — https://developers.cloudflare.com/ai-gateway/
Product page — https://www.cloudflare.com/products/ai-gateway/
Terms of service — https://www.cloudflare.com/terms/
Repository cloudflare/ai — https://github.com/cloudflare/ai
Pricing — https://developers.cloudflare.com/ai-gateway/reference/pricing/
Data Localization — https://developers.cloudflare.com/data-localization/compatibility/
Trust Hub — https://www.cloudflare.com/trust-hub/
SCIM — https://developers.cloudflare.com/cloudflare-one/team-and-resources/users/scim/
Spend limits — https://developers.cloudflare.com/ai-gateway/features/spend-limits/
Analytics — https://developers.cloudflare.com/ai-gateway/observability/analytics/
Coding agents — https://developers.cloudflare.com/ai-gateway/integrations/coding-agents/
Claude Code — https://developers.cloudflare.com/ai-gateway/integrations/coding-agents/claude-code/
Chat completions — https://developers.cloudflare.com/ai-gateway/usage/chat-completion/
Guardrails — https://developers.cloudflare.com/ai-gateway/features/guardrails/
DLP — https://developers.cloudflare.com/ai-gateway/features/dlp/
Dynamic Routing — https://developers.cloudflare.com/ai-gateway/features/dynamic-routing/
BYOK — https://developers.cloudflare.com/ai-gateway/configuration/bring-your-own-keys/

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.

AI Gateway vs. Cloudflare AI Gateway: Global edge service or infrastructure component in your own tenant?