Deploy agents,
pay per token
Build multi-step AI agents with automatic model selection per step. $0.50 per 1M tokens. Provider costs at cost. No markup.
+ provider costs passed through at cost
- Up to 5 agents with multi-step workflows
- 35+ models auto-selected per step
- Streaming, function calling, tool use
- Built-in retries and fallbacks
- In-dashboard test panel
- OpenAI-compatible API
400M tokens included · $0.25/1M overage
- Everything in Agent Starter
- Unlimited agents
- 400M tokens included ($200 value)
- 50% discount on overage tokens
- Custom response memories (up to 10)
- Train routing on your request logs
- 10 training runs included
- Higher rate limits
- Priority support
What's included with every agent
Each agent is a deployable workflow — no infra to manage.
Multi-step workflows
Chain reasoning, execution, and synthesis steps in a single agent.
Smart model selection
Each step auto-picks from 35+ models based on complexity and cost.
Tools & webhooks
Give agents custom tools with webhook endpoints for real-world actions.
OpenAI-compatible
Use model: "agent:your-id" with any OpenAI SDK. Two lines to integrate.
Built for production
Enterprise-grade infrastructure under the hood.
100ms routing overhead
Model selection per step happens in under 50ms. Total latency is routing + model inference.
Data isolation
Agent configs, response memories, and training data are fully isolated per account.
Drop-in API
Works with any OpenAI client — LangChain, CrewAI, AutoGen, Vercel AI SDK. Change baseURL, keep your code.
Deploy your first agent in minutes.
Free tier includes $5 in credits. No credit card required.