Base URL switch
No SDK rewrite
Inference Spend Autopilot
Keep your stack. Cut token spend 35%–85% with automatic model routing.
CrossCompute is an OpenAI-compatible gateway for teams and autonomous agents. Route each request to the right model, enforce quota in credits, and let agents self-top-up over x402 on Base.
Same `/v1/chat/completions` interface
Routing profiles: `cheap` · `medium` · `premium` · `auto`
Agent can buy its own credits via x402
Routing decisions
Per request in milliseconds
Agent payment rail
x402 + USDC on Base
Why this wins
One platform for cost control, quality routing, and autonomous usage monetization.
Spend-aware model routing
Instead of pinning one expensive model for all prompts, CrossCompute routes simple prompts to cheap models and reserves premium models for hard prompts.
Agent self-funding loop
Agents can monitor credits, trigger `/api/agents/topup`, complete x402 payment, and continue inference without manual ops.
Audit-ready billing surface
Every call returns routing and billing headers, so teams and agents can verify model, price rule, and charged credits in real time.
Agent Loop
Design for autonomous agents, not just human dashboards.
Discover
Agent reads `/agent-skill.md`, `/openapi.json`, and `/api-parameters.md`.
Onboard
Agent registers and receives `cc_agent_*` key scoped to tracked credits.
Self top-up
Agent pays stablecoin via x402 challenge-response and gets quota instantly.
Infer + monitor
Agent calls `/v1/chat/completions`, then checks usage and tops up again when low.
Evidence
Routing savings are backed by published research and production router data.
Up to 98%
FrugalGPT reports up to 98% cost reduction at matched top-model quality in benchmark settings.
arXiv: FrugalGPT2x+
RouteLLM reports more than 2x cost reduction while maintaining response quality.
arXiv: RouteLLM30%–70%
Public routing benchmarks commonly show major cost cuts at similar quality targets.
RouteLLM benchmarks100x claim
NotDiamond reports large production improvements in cost-performance for routing workloads.
NotDiamondPricing references: OpenAI, Anthropic, Google Gemini, DeepSeek. Numbers vary over time and by region, caching, and volume discounts.
Machine-readable Surface
Everything external agents need to index, verify, and integrate quickly.
The core promise: less token waste, stronger quality control, and agent-native monetization.
Switch one base URL, keep your OpenAI-compatible calls, and let CrossCompute optimize cost on every request.