Enterprise AI gateway

An enterprise AI gateway built for agents

Leanroute is a managed AI gateway: one API in front of every model your organization approves, with a key per agent, spend caps enforced before each call, guardrails on prompts and responses, MCP tools only on approved servers, and a record of every call. Nothing to deploy or keep running.

One API, every approved model

OpenAI-compatible and Anthropic-compatible endpoints in front of 100+ models from 14 providers. Bring your own provider keys, or use ours. Allow or block providers for the whole organization.

Model directory

A key per agent, revocable on its own

Runtime keys call models and can't manage keys, billing or provider credentials. Admin keys are separate. Owners, admins and members have different rights in the dashboard.

Agent control

Spend caps enforced before the call

Daily and monthly caps per key and per organization, rate limits in requests and tokens per minute, and a budget per agent run that stops a runaway task on its own.

AI cost control

Guardrails on prompts and responses

Block or redact personal data, secrets and prompt injection at the gateway, from a built-in library or your own patterns, so every team gets the same rules.

Guardrails

MCP tools, only on approved servers

Agents use MCP tools on any model, but only on servers on your allowlist, and each key can be limited to some of them. Every tool call is logged.

MCP gateway

A record of every call

Each request is logged with key, agent, run, model, provider, tokens, cost and status. Metadata only: prompt and response content is not stored.

Agent runs

An audit log of every change

Who created or revoked a key, changed a cap, a guardrail, a provider key, the MCP allowlist or the team, and when, with before and after values. Export it as CSV. Secrets are never recorded.

Audit log

A good fit when

  • Several teams or agents share model access and you need one place to set limits and see costs.
  • You want provider contracts and keys to stay yours (BYOK) without running a gateway yourself.
  • Agents call tools, and you need to decide which ones and keep a record of what they did.
  • Data residency in Singapore matters, or you serve customers across Asia.

Where your data goes

The gateway, database and cache run in Singapore. Provider keys you bring are encrypted at rest. Prompts pass through to the provider you chose and are not stored by Leanroute. Sub-processors, residency and the regulations we map to are on the compliance page and the security page.

Not there yet

Large enterprises often require these. We don't have them today:

  • SSO (SAML or OIDC) and SCIM provisioning
  • A SOC 2 report. Automated security scans run monthly; see the security page
  • A self-hosted edition: Leanroute is managed only today

Questions

What is an enterprise AI gateway?
A single control point between your applications or agents and the AI model providers. It gives every team one API, enforces who may use which models, caps spend, applies data rules such as PII redaction, and keeps a record of every call, instead of each team wiring providers in on its own.
How is an AI gateway different from an API gateway?
An API gateway routes and secures HTTP traffic. An AI gateway understands model calls: it counts tokens and cost, enforces budgets in dollars, fails over between model providers, checks prompts and responses, and runs or governs tool calls for agents.
Does Leanroute store our prompts?
No. The gateway logs metadata (key, model, tokens, cost, status, tool names and timings) and never stores prompt or response content, or MCP tool arguments and results.
Is Leanroute ready for a large enterprise?
It depends on your requirements. The controls above are live today. SSO, SCIM, a SOC 2 report and a self-hosted edition are not available yet. If any of those is a hard requirement, tell us; otherwise, compare against Kong, Portkey or LiteLLM Enterprise.

Try it on your own traffic in a few minutes, or tell us what your organization needs.