arb402
AI inference powered by on-chain micropayments
About
The problem it solves 1. Overpaying for AI Inference The Problem: SMEs, indie devs, and early teams pay far more than they consume. They lack optimization expertise (batching, scaling, quantization, cold-start fixes), making small-scale inference extremely expensive. The Solution: We provide an optimized inference layer that performs all cost-reduction techniques behind the scenes. Users get much cheaper inference without needing infra knowledge, and only pay for what they use. 2. Subscription-Based Model Limitations The Problem: Users are forced into monthly subscriptions even with low usage. They pay for idle compute, with no real pay-per-use option. The Solution: We offer true per-request billing. Every request is individually priced, with no monthly commitments or unused credits. Perfect for SMEs with unpredictable or low-volume workloads. 3. Centralized Payment & API Key Management The Problem: Traditional AI platforms rely on: API keys centralized identity centralized payment processing This exposes sensitive information and forces users to trust a closed system. The Solution We use wallet-based authentication — no API keys needed. Payments are transparent, verifiable, and on-chain, removing the need to trust any centralized provider. 4. Gas Cost Barrier for Micro-Payments The Problem: Micro-payments (<$0.001 per request) are impossible on most chains because gas fees are higher than the inference itself. The Solution: We implement gasless micro-payments by batching and settling on-chain efficiently. Users pay near-zero gas, enabling real micro-transactions for each model request. 5. Lack of a Decentralized AI Marketplace The Problem: There is no open marketplace where: model providers can list models users can pay per request usage is trustlessly tracked settlements happen automatically The Solution: We provide a decentralized marketplace for AI models. Users can select any model (LLMs, embeddings, vision models). Providers earn directly and transparently. All usage and payments are recorded on-chain. 6. Opaque Pricing Models The Problem: Today’s pricing is unclear: users can’t verify their usage prices don’t correlate with compute billing is non-transparent The Solution: We offer transparent, auditable pricing: Every request is logged immutably, linking: tokens used → cost → on-chain settlement.
How it's made
Challenges we ran into 1. Making Pay-Per-Request Efficient The biggest challenge was ensuring that each inference call is cheap, predictable, and doesn’t create unnecessary on-chain overhead. How I approached it: Researched different metering patterns, measured gas usage per model type, and optimizing the flow to minimize writes and reduce the cost per request.