AgentRail Spend Gateway

The sub-cent financial firewall and budget coordinator for autonomous AI agents and multi-agent swarms.

Quickstart (3 min)Zero-Overshoot Guarantee

Why AgentRail?

When autonomous agents run in parallel (Fan-Out workflows, swarm collaboration, research loops), traditional cloud gateways fail to enforce budgets during concurrent bursts. AgentRail enforces atomic, pre-flight budget guarantees at the edge.

Zero-Overshoot Guarantee

Atomic pre-flight reservations in SQLite Durable Objects guarantee that spending never exceeds your hard budget (0.00% overrun).

Smart Top-Up (HITL)

Halts processes in an INSUFFICIENT_FUNDS state without terminating the agent thread, allowing operators to increase budget in flight.

Edge Speed (<10ms Overhead)

Deployed across Cloudflare Workers globally. Ingress admission checks complete in single-digit milliseconds before touching LLM networks.

Drop-In Proxy Compatibility

Fully compatible with standard OpenAI, Anthropic, and OpenRouter SDKs. No custom client libraries needed — just override base_url.

Empirical Benchmark: Cloudflare vs AgentRail

In an empirical test under identical $0.0500 USD budget limits and a 180-request concurrency burst:

Metric	Cloudflare AI Gateway	AgentRail Gateway
Configured Budget	`$0.0500 USD`	`$0.0500 USD`
Concurrent Requests	180 requests	180 requests
Admitted (Dispatched)	180 (100.0%)	50 (27.8%)
Blocked at Ingress	0 (delayed cutoff)	130 (72.2%)
Actual Incurred Spend	`$0.1034 USD`	`$0.0500 USD`
Budget Overshoot	+106.8% OVERRUN	0.00% (Zero-Overshoot)
Enforcement Mechanism	Post-Hoc / Eventual	Pre-Flight ACID