I was bored so i built Graze, your new favorite ai agent. Graze(usegraze.world) is a platform that supplies cheap AI compute for agents (especially those active on X/Twitter) by automatically routing each call to the cheapest model capable of handling it. It claims roughly 93% savings versus Claude Sonnet 4.6 list prices on typical agent workloads, and it lets anyone tip AI compute directly to an X handle. Core features includes Smart cheap routing — Uses graze/auto, which starts with a fast/cheap model (e.g. llama-4-maverick) and fails over to stronger ones only when needed. Measured real calls (≈1,700 input + 160 output tokens) cost about $0.49 per 1,000 calls versus ≈$7.40 at Claude list price (≈93% cheaper), with a 5% Graze fee included. Failed attempts cost nothing.⁠Usegraze Send compute to anyone on X: Pick a handle and an amount ($0.50–$25), share a tip on X, and the recipient claims it simply by logging in with X. Unclaimed tips return after 30 days. No wallet or extra forms required.⁠Usegraze One-line integration: Point any OpenAI-compatible agent (OpenAI SDK, LangChain, Hermes, etc.) at Graze’s base URL and set the model to graze/auto. Credits live in Graze; agents spend them without needing separate accounts at each lab. Every response includes a receipt showing model used, cost, and savings.⁠Usegraze Safety and transparency: Pre-call holds prevent balances from going negative. Beta balances are capped at $50 per agent. A public live “field” dashboard shows real calls and savings in real time (visualized as growing grass blades). In short, Graze is positioned as low-cost, X-native AI compute infrastructure for agents, with transparent routing, easy tipping of credits, and token mechanics tied to usage volume.