Takusen overview

Takusen turns questions that ordinary smart contracts cannot answer into typed, verifiable results. An application asks a bounded question, chooses one or more AI models and an output schema, and prepays a maximum in USDC. Takusen coordinates independent model executions, verifies how each answer was produced, and publishes a result that applications can read through the API or settlement contracts.

Ask opens the judgement workspace without trapping the landing page: before the first valid question is submitted, scrolling upward returns to the product introduction. After submission, the outer page stays anchored while the conversation thread scrolls independently. The workspace menu's Home action always releases that anchor and returns to the landing page.

• Input integrity: the API, enclave, observers, and receipts share the same canonical request commitment.
• Bounded spend: a request reserves a caller-approved maximum; unused reservation is released at finalization.
• Typed settlement: the selected domain stores compact schema-bound outputs rather than treating free-form prose as the result.
• Attributable execution: a receipt binds model, provider, usage, cost, the consensus provider-cost ceiling, exchange hashes, archive ID, execution nonce, and TEE key.
• Quorum-authorized settlement: no single API server, model provider, or executor can finalize a payable result by itself.
• Replay resistance: request IDs, execution IDs, batch nonces, epoch markers, burn IDs, and one-shot challenges each protect their own boundary.