What LLM Gateway does and how it earns

What LLM Gateway does and who it is for

LLM Gateway is an open-source API gateway that routes requests across more than 250 models from more than 40 providers, including OpenAI, Anthropic and Google S1. Developers keep their existing OpenAI SDK code and change only the base URL to start using it S1. The gateway tracks costs in real time and lets teams switch providers without changing application code S1.

The product is built for teams that are tired of juggling API keys and provider dashboards S1. It gives them one dashboard for all provider keys, per-model and per-provider spend breakdowns, error and reliability monitoring, and a project-level usage explorer S1. Spend caps can be set on any API key or member, and requests stop once the cap is reached S1. When a provider errors, times out or rate-limits a request, the gateway retries up to twice on another provider for that model within the same request S1.

Larger organisations get a separate tier of features. Enterprise adds SSO through Okta, Entra ID or another SAML 2.0 provider with SCIM provisioning, roles per organisation and project, audit logs, and guardrails that detect emails, card numbers, secrets, prompt injection and jailbreak attempts before a provider sees them S1. Enterprise customers can also restrict routing to providers that meet rules on headquarters country, data retention and training S1. The company says it is SOC 2 Type II compliant and lists Samsung, Harvard, Coloop.ai and FieldKo as customers S1.

The gateway runs either in LLM Gateway's cloud or on a customer's own infrastructure with Docker or Kubernetes using a Helm chart S1. The core is open source under AGPLv3 and the enterprise code is source-available, but enterprise features on a self-hosted install need a license key S1.

How it charges

LLM Gateway has two plans. The Free plan is $0 forever, and the Enterprise plan is custom priced through sales with a 30-day trial S2. There are no seats, no minimums and no subscription S2.

The main charge is usage based. Users pay per token at each provider's own rates, and the only platform fee is a flat 5% when they buy credits S2. Non-US cards pay an extra 1.5% international fee S2. Paid models are not rate limited, while free models are capped at 20 requests per minute on the Free plan S2.

Bringing your own provider keys removes the platform fee entirely. With BYOK, routing through LLM Gateway is free and users pay their providers directly while still getting unified analytics, caching and failover S2. The homepage describes this as free forever with no credit card required S1.

The one optional charge is storage. Enterprise organisations that enable full request and response retention are billed $0.01 per 1M tokens, while everyone else gets metadata-only logs at no cost S2. Enterprise also adds volume discounts on credits, custom routing, 30 days of full payload retention and a 99.9% uptime SLA S2. Refunds are self-serve: if less than 10% of a credit purchase has been used, a user can request a refund from the billing dashboard within 14 days of buying S2. Self-hosting the open-source gateway is free S2.

What it earns

LLM Gateway reports monthly recurring revenue of $78,764 as of 2026-10-22 S3. Agent Made gives this figure a grade A, which means it is Stripe-verified on TrustMRR S3.

Our take: why it works

This section is opinion.

The free BYOK tier removes the main reason a developer would hesitate to try a gateway. Nothing is charged until a team decides it would rather buy credits through one bill than manage a dozen provider accounts, and at that point the 5% fee is small relative to the convenience of unified spend tracking and failover.

The pricing also ties revenue to customer usage rather than to seat counts, so revenue grows as customers ship more. The Enterprise tier then captures the organisations that need SSO, audit logs and compliance policies, which are exactly the features a procurement team will insist on before an AI product goes to production.

The AGPLv3 core gives security-conscious buyers a way to inspect the code and self-host, while the license key for enterprise features keeps a reason to pay. That combination lets the open-source project act as a funnel for the hosted and licensed product.

Our take: how builders can work with it

This section is opinion.

• Use it as the model layer. A builder shipping an agent or SaaS product can point their OpenAI-compatible client at the gateway and gain provider switching, spend caps and retries without writing that infrastructure themselves.
• Build around the analytics. Per-project usage and cost data could feed a builder's own billing, so customers of a wrapper product are charged in line with what the gateway reports.
• Contribute to the open-source core. The AGPLv3 repository is a place to add providers, integrations or guardrail rules that a builder needs, and the company lists integrations, templates and an AI SDK provider as entry points.
• Self-host for a niche. A builder serving a regulated market could run the gateway inside a customer's network and offer their own product on top, paying for an enterprise license only when those features are needed.

Related open tools

• opencodex: Provider proxy so Codex and Claude Code can run on any LLM S6.
• treg: One MCP server and key for 4,000+ paid data tools, pay per call S7.
• XERJ.ai: Search engine for agents, one Rust binary for search, vectors and memory S5.
• Trail-code: Self-learning shared memory for coding agents stored in your repo S4.
• skills.sh: Directory and installer for agent skills S8.