Krump x UCP MVP — agentic creator economyKrump Protocol Agents is a UCP-native demo where AI sub-agents orchestrate tips, paid tutorials, and live battles — settled in USDC on Arc Testnet under a Vyper-mirrored policy contract.
Every action — tipping, unlocking, booking, bounty payout — is a UCP checkout with the same identity, the same policy, the same receipts.
Live read of /api/config. Toggling environment variables flips Circle, MetaMask, and the Vyper-mirrored settlement on or off.
Run an action to see output.
ENS gives every agent a portable, on-chain identity. The Vyper-mirrored policy spends against that identity — so an agent can only transact on the intents its ENS authorizes, up to the policy caps, settled in USDC on Arc.
Each agent is bound to an ENS name on Sepolia. Text records carry agentId, capabilitiesUri, allowedIntents, and the Arc actor address — resolved live with viem.
runSession enforces allowedIntents from ENS text records, and for high-risk intents (challenge_payout, crew_split_settlement) requires a non-empty ENSIP-25 attestation at agent-registration[<registry>][<agentId>] — a registry-anchored proof of agent registration.
Once gating passes, the Vyper-mirrored policy enforces max_ticket_minor and daily_cap_minor per agent_id, with proofs stamped with the settlement contract + identity registry.
ENS name (Sepolia) | text records: agentId, allowedIntents, arcAddress | + ENSIP-25 text["agent-registration[<registry>][<agentId>]"] v runSession(intent) --> intent in allowedIntents? --> no --> reject | yes v high-risk intent? --> yes --> ENSIP-25 verified? --> no --> reject | no / yes-and-verified v Vyper policy (max_ticket, daily_cap per agent_id) | approved v UCP checkout --> Circle USDC settlement on Arc | v AIsa LLM --> api_key_proxy chat | x402 probe | x402 external-settle (pay-per-call compute)
AIsa closes the loop: the same identity + policy layer that gates commerce also meters LLM compute. Run real multi-turn chat throughapi_key_proxy, or exercise the x402 payment-required protocol with probe and external-settle modes.
Agents shouldn't hold private keys. KeeperHub gives our agents an org-bound executor with policy controls, spending caps, and cross-chain reach — so the Vyper-mirrored UCP policy can authorize a payout and KeeperHub carries it out, on Arc or anywhere CCTP reaches.
/execute/transfer on Arc Testnet — no user signing, org-key gated, settled in a single round-trip.
Arc → Base, Polygon, Arbitrum, Avalanche, or Ethereum Sepolia via CCTP burn + attest + mint, then a native USDC payout on the destination.
Org-level signer, spending caps, allowlists, observable execution IDs, and a built-in /diagnose endpoint for 403s.
agent intent --> UCP policy check --> KeeperHub /execute/transfer
|-- local: Arc Testnet (no bridge)
`-- online: CCTP burn -> attest -> mint -> KeeperHub payoutAIsa is the LLM rail for Krump agents. One OpenAI-compatible endpoint (/v1/chat/completions) fronts 60+ frontier models across text, coding, vision, image, audio, and video — billed per call. The same ENS-gated, Vyper-policied identity that authorizes a tip or a battle payout also meters AIsa compute.
gpt-5.2, claude-opus-4-7, gemini-3.1-pro-preview, qwen3-max, deepseek-v3.2, kimi-k2.5, MiniMax-M2.5.
gpt-5.3-codex, qwen3-coder-plus, claude-sonnet-4-6-thinking, deepseek-v3.2.
qwen3-vl-plus, gemini-2.5-pro, claude-opus-4-7, seed-2-0-pro.
wan2.7-image-pro, seedream-4-5, gemini-3-pro-image-preview.
Speech + audio-capable Gemini and Qwen tiers via the same /v1/chat/completions surface.
Video-capable seed and wan tiers, metered per call.
Real multi-turn chat. Server holds the AIsa key; client streams tokens via SSE.
Force a 402 Payment Required envelope so agents can negotiate price before paying.
Replay the request with settlement headers — proves end-to-end pay-per-call flow.
Clients hit /api/ucp/discovery to learn capabilities and the REST endpoint base.
Pick an intent (tip, unlock, battle). Sub-agents propose; the payments agent builds a UCP checkout.
Off-chain demo, MetaMask, or Circle USDC transfer — gated by a Vyper-mirrored policy.
Start with a tip, unlock a clip, or open the agent orchestrator and watch the trace fill in.