LEANROUTE

Singapore-hosted AI Gateway · MCP on any model

LLM routing and MCP tools, on any model.

We run the agent loop at the gateway, so the same MCP tools work on GPT, Gemini, DeepSeek and Groq, on the standard OpenAI API, fully hosted.

100+ models behind one OpenAI-compatible endpoint. Or point Claude Code at us and cut that bill roughly tenfold.

Claude Code in one env varPer-org MCP allowlistsZero per-request markup
Get an API key →Try the playground, no signup

100 free requests when you sign up. Fast, low-cost models. No card, no provider keys.

OpenAI-compatible. Point your SDK at api.leanroute.dev/v1 and you're running.

60 seconds to first requestsign up → mint key → change one env var → done
LEANROUTEapi.leanroute.dev · Singapore-hostedcache hit · $0no provider callYour appSDK or agentAPISOpenAI API+ mcp_serversAnthropic APIClaude CodeCONTROLSSpend capsorg + key, before dispatchGuardrails7 rules, block or redactSemantic cacheper-org, hit = $0Routingtier pick + auto-failoverLLM PROVIDERSOpenAIAnthropicGoogleDeepSeek130+ models · 14 providersMCP SERVERSyour Calendaryour Docsany MCP URLtools run on any model5xxfailover$0
130+ models
all responding
Across 14 curated providers. Retired ids are rewritten, not rejected.
0
customer-visible incidents
100.00% uptime since launch. Every incident logged.
17
auditable response headers
What every call proves, documented.

The chat bubble in the corner is built on Leanroute + RAG. Read the writeup →

Platform

Five pillars. One control plane.

Everything below is shipping in production today.

LLM routing

100+ models behind one OpenAI-compatible endpoint, curated across 14 providers. Cheap-fast tier, smart routing, failover on upstream 5xx.

Model catalog

Prompt & semantic cache

Semantic cache serves near-duplicate prompts at $0, org-isolated. Exact matches free too.

How caching works

Guardrails

Regex block, regex redact, and OpenAI moderation. Seven-rule starter library ships in.

Rule library

MCP on any model

MCP tools on GPT, Gemini, DeepSeek and Groq, not just Claude, because the gateway runs the agent loop itself. Per-org allowlists; chain-aware agent policy on the roadmap.

MCP docs

Spend control

Hard daily and monthly caps per org and per key. Rate limits enforced at the edge.

Pricing

Dashboard

Every dollar, every provider, every request.

Cost by provider over time. Cache-hit rate. Per-request logs with model, latency, and status. Included on every plan.

leanroute.dev/dashboard/usage
Leanroute usage dashboard: request, token, cost and cache-hit cards, a cost-by-provider line chart, and a per-provider table covering Anthropic, DeepSeek, xAI, Kimi, Google, OpenAI, Sarvam, Groq, Qwen and Krutrim with real spend data since June

Real screenshot from a production account. See the full walkthrough at /docs.

Providers

100+ models, one endpoint, your existing SDK. Curated across 14 providers rather than scraped from every host with an API.

Every provider below is reachable through api.leanroute.dev/v1. Change the base URL, keep everything else. Full model catalog at /docs/models.

OpenAI
Anthropic
Google
xAI
DeepSeek
Qwen
GLM
Doubao
Kimi
Sarvam
Krutrim
Meta
Groq
AWS Bedrock

Pricing

Flat monthly BYOK or pay-as-you-go credits.

Zero per-request markup on inference either way. Every new account starts with 100 free requests on fast, low-cost models, no card needed.

Starter

$15/mo

1M requests · 30 RPM

Pro

Recommended
$25/mo

5M requests + $3/1M overage · 150 RPM

14 days free · no card needed

Credits

5% platform fee

Applied at deposit only · $0 per request · $50 min top-up

Larger team?

Let's talk

Design-partner terms, email us

At $1,000/mo of provider spend:Leanroute Pro $1,025vs OpenRouter $1,050vs Portkey Prod $1,049+

Zero per-request markup means the savings widen as your bill grows. Run your own numbers →

Self-hosted / private-deploy? Not shipped yet. We're gauging demand.Join early-access list →
Read the full FAQ →See how we compare →