Drop in. Route smart. Cut costs.
A drop-in replacement for the OpenAI and Anthropic base URLs. Pick a section, or start with the three-line quickstart.
Most agent loops pin one expensive model across every step, from planning to retrieval to tool calls to synthesis to verification, and pay frontier prices throughout. IQ Routing classifies each step and sends it to the cheapest model that still holds quality, then serves repeats from cache instead of billing you twice. The same session carries an audit trail of every model call, a capability resolver so you name intent instead of pinning models, and a cost envelope that caps spend before a runaway loop blows the budget.
Quickstart
Switch the OpenAI base URL to the gateway in three lines.
Open →ReferenceAPI reference
Auto-generated from the gateway's OpenAPI schema.
Open →TemplatesSDK recipes
Runnable templates for OpenAI, Anthropic, curl, LangChain, Cursor, Codex CLI, and Claude Code.
Open →Core concepts
Auth
How API keys, scopes, and the gw_live_ prefix work.
Routing
How the gateway picks a model and a none/low/medium/high thinking budget per prompt.
Step classifier
Per-step labels that bias routing across an agent loop.
Capability aliases
Name intent with cap: handles instead of pinning models.
Agent sessions
One row per agent run: trajectory replay and cost envelopes.
Caching
Exact, semantic, multi-turn, and streaming cache hits, billed at zero.
Billing
Token classes, four-way spend split, and thinking-budget reporting.
Tooling
Security & compliance
Operations
Three lines today. A clean bill tomorrow.
No charge to start. Card on file, no trial. You'll route real traffic in under five minutes.