Smart routing
Each request is matched by intent, load, and latency budget to the right model and provider. No more hardcoded failover chains.
AGENT ROUTING, DONE RIGHT
Every agent call, through one gateway
Agent Gateway sits between your models and your tools, so each request lands in the right place: a fast triage for easy calls, deep reasoning for the hard ones. Auth, rate limits, and traces come along for free.
Hono on Bun · LangGraph orchestration
OpenAI, Anthropic, Gemini, local models
ag — agent gateway
$→→→→→✓
- 99.99%
- uptime over 12 months
- <25ms
- median triage time
- 12B+
- requests routed
- 120+
- models and tools
01 / FEATURES
Production plumbing, without the plumbing
Everything an agent team ends up hand-rolling in the second month, shipped as a single, auditable layer.
Auth and secrets
Model keys never touch your backend. Issue per-call credentials and rotate secrets from a single vault.
Observability
Every call, token, and dollar traceable. Spot a slow model before your customers do, and replay any request.
Tool integrations
Native connectors for SQL, HTTP, and MCP tools with retries, timeouts, and schema validation built in.
Rate limiting
Per-key, per-team, and per-model budgets that fail fast at the edge instead of surprise-ing you on the bill.
Webhooks
Fire-and-forget callbacks on completion, failure, and budget events straight into your own pipeline.
02 / HOW IT WORKS
Two speeds of thought, one coherent flow
A cheap triage decides what matters, and only the genuinely hard work spends real compute. The whole graph is orchestrated server-side.
01
INGEST
One endpoint
Everything enters through POST /v1/route, behind enforced auth. Streaming and tool calls stay on the same wire.
02
SYSTEM 1 · TRIAGE
Fast classification
Easy requests are classified and routed in milliseconds: intent, language, and tool selection with no heavy reasoning.
03
SYSTEM 2 · REASONING
Deep planning
Complex, multi-step tasks hand off to LangGraph, which builds a plan, calls tools, and steers sub-agents to a result.
The runtime
Hono on Bun · LangGraph orchestration
honobunlanggraphstreamedge-ready
03 / PRICING
Pay for what you route
Start free. Scale when your agents do. Every tier includes auth, rate limits, and 7-day trace retention.
Hobby
For side projects and early prototypes.
$0/ month
- 50k requests / month
- 1 project
- 3 models
- Community support
Pro
For product teams shipping in production.
$49/ month
- 2M requests / month
- Unlimited projects
- Advanced routing + caching
- 60-day trace retention
- Priority support
Scale
For platforms and high-volume workloads.
$199/ month
- 20M requests / month
- Custom providers
- SSO and audit logs
- Dedicated routing node
04 / FAQ
Questions, answered
Anything else, find us in the community.
How fast can I get my first agent routed?
Add a key and a model, point your SDK at the gateway, and you are routing inside about ten minutes. No infrastructure to stand up.
What is the difference between systems 1 and 2?
System 1 is a fast triage that classifies easy requests and routes them straight to the right model or tool. System 2 is a deeper LangGraph planner for multi-step tasks that need reasoning across several calls.
Which models and providers do you support?
OpenAI, Anthropic, Gemini, and any OpenAI-compatible endpoint, including self-hosted and local models. Custom providers are available on the Scale tier.
Where do my secrets live?
In our vault, encrypted at rest, and never shipped to your backend. The gateway issues scoped, per-call credentials to your tools.
Do you support on-prem or VPC deployments?
Yes. The Scale tier can run a dedicated routing node inside your own VPC, so requests and traces never leave your network.
What happens if I go over my request limit?
Requests fail fast with a clear 429 and a header telling you which budget it hit. No surprise bills, and alerts can fire before you reach the edge.
Put your whole stack behind one door
Route your first request today. No credit card, no model keys in your repo, no lock-in.
Start routing free