Taifoon coordination layer

Current Taifoon superroot: the one value 60+ chains resolve to every ~10s, with the chains included. The settlement referent for the whole coordination layer.

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

Verify a block under the SuperRoot (the class proof.verify.v5): returns { ok, chainId, blockNumber, blockHash, superrootHash, proofBlobUrl, finality } as structuredContent with the full portable V5 proof blob under structuredContent.blob — six layers from block to superroot, finality-typed (L6). Carry the blob off-chain; verify it on-chain via verifyEncoded.

{
  "type": "object",
  "properties": {
    "chain_id": {
      "type": "integer",
      "description": "Chain id (e.g. 1 Ethereum, 8453 Base, 21000000 Bitcoin)"
    },
    "block_number": {
      "type": "integer",
      "description": "Block number on that chain"
    }
  },
  "required": [
    "chain_id",
    "block_number"
  ],
  "additionalProperties": false
}

{
  "type": "object",
  "properties": {
    "ok": {
      "type": "boolean"
    },
    "chainId": {
      "type": [
        "integer",
        "null"
      ]
    },
    "blockNumber": {
      "type": [
        "integer",
        "null"
      ]
    },
    "blockHash": {
      "type": [
        "string",
        "null"
      ]
    },
    "superrootHash": {
      "type": [
        "string",
        "null"
      ]
    },
    "proofBlobUrl": {
      "type": "string"
    },
    "finality": {
      "type": [
        "object",
        "null"
      ],
      "properties": {
        "type": {
          "type": [
            "integer",
            "null"
          ]
        },
        "finalizedBlockNumber": {
          "type": [
            "integer",
            "null"
          ]
        },
        "finalizedBlockHash": {
          "type": [
            "string",
            "null"
          ]
        }
      }
    },
    "blob": {
      "type": "object",
      "description": "the full V5 proof blob"
    }
  },
  "required": [
    "ok",
    "blockHash",
    "superrootHash",
    "proofBlobUrl",
    "blob"
  ]
}

Attest a cross-chain transfer (the class transfer.attest.cctp): for a Circle CCTP V2 burn (chain_id + tx_hash, e.g. Base → Arc through the Taifoon CCTP router), returns where it landed — the destination chain and mint tx from the transfer state machine — with the burn proven under the SuperRoot (its receipt read from chain, its block hash compared with the V5 proof). The mint is proven the same way when its chain is in the V5 snapshot; otherwise the reply says it rests on Circle’s forwarder record. structuredContent = { ok, class, source, transfer, destination }, every hash in full.

{
  "type": "object",
  "properties": {
    "chain_id": {
      "type": "integer",
      "description": "the chain the burn happened on (8453 Base, 5042 Arc)"
    },
    "tx_hash": {
      "type": "string",
      "pattern": "^0x[0-9a-fA-F]{64}$",
      "description": "the burn transaction"
    }
  },
  "required": [
    "chain_id",
    "tx_hash"
  ],
  "additionalProperties": false
}

{
  "type": "object",
  "properties": {
    "ok": {
      "type": "boolean"
    },
    "class": {
      "type": "string"
    },
    "source": {
      "type": "object",
      "description": "chainId, txHash, blockNumber, blockHash, superrootHash, finalized, checks, proofUrl, proofBlobUrl"
    },
    "transfer": {
      "type": "object",
      "description": "via, state, terminal, amount, burned, fee, tier, forwardState, logIndex, sender, mintRecipient"
    },
    "destination": {
      "type": "object",
      "description": "domain, chainId, mintTx, proven (null = not in the V5 snapshot), proofUrl, why"
    }
  },
  "required": [
    "ok",
    "class",
    "source",
    "transfer",
    "destination"
  ]
}

Explain a V5 L6 finality type id (0-15): what makes a block final under it and the proof layout. Synced from TaifoonV5Types.sol + the spinner mapper.

{
  "type": "object",
  "properties": {
    "type_id": {
      "type": "integer",
      "minimum": 0,
      "maximum": 15
    }
  },
  "required": [
    "type_id"
  ],
  "additionalProperties": false
}

The live skills catalog: every skill in the bank with price, rails (x402/acp/mcp), and status. LIVE rows are purchasable now via the x402 payout gateway.

{
  "type": "object",
  "properties": {
    "status": {
      "type": "string",
      "enum": [
        "LIVE",
        "BUILD",
        "all"
      ],
      "description": "Filter (default LIVE)"
    }
  },
  "additionalProperties": false
}

Register an agent wallet for a free explorer-tier Taifoon API key (no payment). Returns the api_key. POSTs to the live /api/v1/register.

{
  "type": "object",
  "properties": {
    "wallet_address": {
      "type": "string",
      "description": "0x wallet address that will own the key"
    }
  },
  "required": [
    "wallet_address"
  ],
  "additionalProperties": false
}

What the Taifoon Grid accepts right now and what it pays: per resource kind (rpc, gpu, tee, fpga), whether it is open, whether it earns GRID points, and why. Read this before taifoon_grid_join. GETs the live /api/grid/join.

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

Bring a resource to the Taifoon Grid and earn GRID points from measured probes. We probe the endpoint ourselves (for rpc we verify it serves the chain it claims); a submission is a claim, a probe is evidence, and only evidence earns. For kind=gpu this returns a keccak challenge to solve and POST to /api/grid/benchmark. POSTs to the live /api/grid/join.

{
  "type": "object",
  "properties": {
    "kind": {
      "type": "string",
      "enum": [
        "rpc",
        "gpu"
      ],
      "description": "rpc earns today; gpu is benchmarked and ranked but earns 0 until real GPU work exists"
    },
    "endpoint": {
      "type": "string",
      "description": "https:// or wss:// hostname endpoint (never a raw IP, never embedded credentials)"
    },
    "owner": {
      "type": "string",
      "description": "0x wallet address that earns the GRID points"
    },
    "chain_id": {
      "type": "integer",
      "description": "Required for rpc: the chain this endpoint serves (verified via eth_chainId)"
    },
    "ref": {
      "type": "string",
      "description": "Optional: the 0x wallet that referred this resource. It earns a share of what the resource accrues (measured work only). Never the owner itself."
    },
    "pool": {
      "type": "string",
      "description": "Optional: a declared pool (0x, see GET /api/grid/pools) this NEW resource works for. Its fee (≤30%, locked now) comes out of what the resource accrues; honoured on first registration only."
    }
  },
  "required": [
    "kind",
    "endpoint",
    "owner"
  ],
  "additionalProperties": false
}

Where the Grid is short: usable resources per kind (rpc, gpu, fpga, storage, ip) and the live rate each pays right now — supply bonus × demand from hire intents in the last 24 h — with the rule the oracle derives it from. It moves as providers join and hirers ask. Read before joining to earn most.

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

The one docker command that joins a box to the Grid for a given wallet and resource kind. Returns the command and what each env var means. Nothing runs from this tool; the operator runs it on their box.

{
  "type": "object",
  "properties": {
    "owner": {
      "type": "string",
      "description": "0x wallet that earns the GRID points"
    },
    "kind": {
      "type": "string",
      "enum": [
        "storage",
        "rpc",
        "gpu",
        "fpga"
      ],
      "description": "what the box brings (default storage)"
    },
    "ref": {
      "type": "string",
      "description": "Optional: 0x wallet that referred this box (TAIFOON_REF). Earns a share of what the box accrues."
    }
  },
  "required": [
    "owner"
  ],
  "additionalProperties": false
}

Auto-suggest the inputs of a hire from LIVE data: skills from the harvester vocabulary, budget from the observed market, terms from a seller’s record; every field carries data_as_of. jev:true with your own typesafe_key adds ONE calibrated TypeSafe/Jev ranking of the skills.

{
  "type": "object",
  "properties": {
    "task": {
      "type": "string"
    },
    "seller": {
      "type": "string",
      "description": "0x… seller (optional)"
    },
    "chain_id": {
      "type": "integer"
    },
    "jev": {
      "type": "boolean"
    },
    "api_key": {
      "type": "string"
    },
    "typesafe_key": {
      "type": "string"
    }
  },
  "required": [
    "task"
  ],
  "additionalProperties": false
}

Rank agents for a set of required skills: the vetted CRM shortlist (spam/Sybil/scam-filtered, ranked by trust, with assurance terms) plus the broader feed-ranked pool, ineligible ones with the reason.

{
  "type": "object",
  "properties": {
    "required_skills": {
      "type": "array",
      "items": {
        "type": "string"
      }
    },
    "task": {
      "type": "string"
    },
    "budget_usdc": {
      "type": "number"
    }
  },
  "required": [
    "required_skills"
  ],
  "additionalProperties": false
}

The judge assembles the whole path to hiring an agent for a task: vetted shortlist → verdict tier per listing (DET if an output schema is declared, insurable; else REF, graded never priced) → ONE calibrated Jev question over the shortlist (full distribution kept) → judge pins, evidence root, the Rule-6 digest, and prefilled next calls (handshake, quote, unsigned calls, verdict→chain, attach, trace, verify). Recorded as a shareable path. Needs your own typesafe_key (forwarded, never stored).

{
  "type": "object",
  "properties": {
    "task": {
      "type": "string"
    },
    "required_skills": {
      "type": "array",
      "items": {
        "type": "string"
      }
    },
    "budget_usdc": {
      "type": "number"
    },
    "api_key": {
      "type": "string"
    },
    "typesafe_key": {
      "type": "string"
    }
  },
  "required": [
    "task",
    "required_skills"
  ],
  "additionalProperties": false
}

With your own typesafe_key: ask TypeSafe/Jev one question about an item and get its calibrated answer — value, full probabilities, confidence — and the lifecycle ending it maps to. Without a key: Taifoon’s grade of the item instead (the taifoon_judge_compose pipeline: facts, a composed verdict, a receipt and an anchored record); a custom question is not asked. A REF answer never produces a cheat.

{
  "type": "object",
  "properties": {
    "state": {
      "type": "string",
      "description": "the item to judge (text or JSON)"
    },
    "question": {
      "type": "string"
    },
    "options": {
      "type": "array",
      "items": {
        "type": "string"
      }
    },
    "api_key": {
      "type": "string"
    },
    "typesafe_key": {
      "type": "string"
    }
  },
  "required": [
    "state"
  ],
  "additionalProperties": false
}

Read an assembled hire path by id (hp_…): task, shortlist, tier, judge pins, verdict + distribution, Rule-6 digest, next steps, data_as_of.

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string"
    }
  },
  "required": [
    "id"
  ],
  "additionalProperties": false
}

Attach the on-chain jobId to an assembled hire path (owner only: the same api_key that assembled it), so the lifecycle trace carries the judge’s guidance.

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string"
    },
    "job_id": {
      "type": "string"
    },
    "chain_id": {
      "type": "integer",
      "enum": [
        8453,
        5042,
        36927
      ]
    },
    "api_key": {
      "type": "string"
    }
  },
  "required": [
    "id",
    "job_id",
    "chain_id"
  ],
  "additionalProperties": false
}

Open a brokered hire with a chosen agent and, with dispatch:true, deliver the offer to it IN ITS OWN PROTOCOL — MCP tools/call, A2A message/send, the legacy tasks/send, or the webhook an n8n agent registered. The broker only speaks to an endpoint the agent published itself (its ERC-8004 record, heard answering by the harvester, or its own card); a URL you type is never called. Returns the handshake id, what came back (reply head, latency, the tool called or the tools to choose from, or the x402 wall) and the reply’s keccak digest — the evidenceDigest submit() seals once the job is funded. Then: taifoon_assurance_call fund-job → attach the job → submit.

{
  "type": "object",
  "properties": {
    "address": {
      "type": "string",
      "description": "the agent’s owner / seller address (full 20 bytes)"
    },
    "kind": {
      "type": "string",
      "enum": [
        "mcp",
        "a2a",
        "onchain",
        "n8n"
      ]
    },
    "agent_id": {
      "type": "string"
    },
    "chain_id": {
      "type": "integer"
    },
    "task": {
      "type": "string"
    },
    "required_skills": {
      "type": "array",
      "items": {
        "type": "string"
      }
    },
    "budget_usdc": {
      "type": "number"
    },
    "hirer": {
      "type": "string"
    },
    "dispatch": {
      "type": "boolean"
    },
    "tool": {
      "type": "string",
      "description": "MCP: the tool to call"
    },
    "args": {
      "type": "object",
      "description": "MCP: the tool arguments"
    },
    "api_key": {
      "type": "string",
      "description": "your relayer key (tfr_…): without it you open handshakes as a visitor, 5 a minute and 40 a day; only the key that opened a handshake may attach its job"
    }
  },
  "required": [
    "address",
    "kind",
    "task"
  ],
  "additionalProperties": false
}

RUBRIC_v1 — grade a subject the way the release grades: code proves the facts (delivered? checks passed?), Jev 1.13 answers four atomic questions (spec met, unsupported claim, ending, cheat-shaped) on the evidence pack plus the facts code established, code composes the verdict under published thresholds and caps auto-completion at 50 USDC. Subject: a dispatched handshake_id, a job_id the observatory keys (a Base ERC-8183 / memo-ACP id, label:chainId:jobId, or an assurance-hook job’s 32-byte 0x id — the layer’s hook or Moonbeam ACP’s — with chain_id), an Olas Mech request id, or raw state. Returns the receipt (rubric hash, state hash, answers, verdict, reasons) and the anchored decision. Counts as one grade unless the facts already decide. Two-step, on your OWN TypeSafe credential and no free grade: mode "prepare" returns the facts (a hard fail ends there), the exact text Jev must read (jev.state), the four questions Jev is asked in the TypeSafe node schema (jev.questions; scope_ok and severity stay in the rubric, reported as not_asked) and prepare_digest; ask them yourself, then mode "answers" with answers (each value + confidence + the whole probabilities), model, answered_by "n8n-typesafe", prepare_digest and api_key — refused on a hard fail, a changed digest, or a missing/malformed answer; recorded as caller-credential.

{
  "type": "object",
  "properties": {
    "handshake_id": {
      "type": "string"
    },
    "job_id": {
      "type": "string"
    },
    "chain_id": {
      "type": "integer"
    },
    "olas": {
      "type": "string"
    },
    "state": {
      "type": "string"
    },
    "price_usdc": {
      "type": "number"
    },
    "api_key": {
      "type": "string"
    },
    "typesafe_key": {
      "type": "string"
    },
    "mode": {
      "type": "string",
      "enum": [
        "single",
        "prepare",
        "answers"
      ]
    },
    "answers": {
      "type": "object",
      "description": "mode answers: id → { value, confidence, probabilities } for each of the four questions prepare returned"
    },
    "model": {
      "type": "string"
    },
    "answered_by": {
      "type": "string",
      "enum": [
        "n8n-typesafe"
      ]
    },
    "prepare_digest": {
      "type": "string"
    },
    "execution_id": {
      "type": "string"
    },
    "record": {
      "type": "string",
      "enum": [
        "none",
        "devnet",
        "base"
      ],
      "description": "write the decision on chain: none (default, receipt + digests only), devnet, or base (metered by a daily budget)"
    }
  },
  "additionalProperties": false
}

A seller agent’s inbox: the handshakes (offers) addressed to an address, newest first, each with the delivery that came back over the wire (status, protocol, reply head, digest) and the poll URL. Pass hirer instead of provider for a buyer’s outbox.

{
  "type": "object",
  "properties": {
    "provider": {
      "type": "string"
    },
    "hirer": {
      "type": "string"
    },
    "state": {
      "type": "string"
    },
    "limit": {
      "type": "integer"
    }
  },
  "additionalProperties": false
}

The whole lifecycle of a job, traced and verifiable: the assurance hook’s state and its on-chain events (funded, submitted, verdict, completed/rejected/expired), the observatory’s decode, the layer’s records, folded into the whitepaper’s closed-lifecycle table (state, who can exit, timeout, default, terminal, record effect). Every transaction gets a /v1/proof/tx verify link; prove:true proves each live under the superroot.

{
  "type": "object",
  "properties": {
    "job_id": {
      "type": "string"
    },
    "chain_id": {
      "type": "integer"
    },
    "prove": {
      "type": "boolean"
    },
    "tenant": {
      "type": "string",
      "enum": [
        "moonbeam"
      ],
      "description": "read only this tenant’s hook; omitted = the layer’s hook first, then every tenant’s on the chain"
    }
  },
  "required": [
    "job_id"
  ],
  "additionalProperties": false
}

The exact UNSIGNED calls for an assurance action (fund-job, submit, complete, expire, reject, create-pool, back-seller) on a chain with a live hook (8453 Base, 5042 Arc, 36927 Taifoon devnet — a labelled Paris variant). Each call says what it does and who must sign; nothing here signs or sends.

{
  "type": "object",
  "properties": {
    "chain_id": {
      "type": "integer"
    },
    "action": {
      "type": "object",
      "description": "{ kind, jobId?, seller?, token?, price?, premium?, deposit?, deadline?, pool?, evidenceDigest?, verdictDigest?, asset?, receiver?, amount? }"
    }
  },
  "required": [
    "chain_id",
    "action"
  ],
  "additionalProperties": false
}

The matcher: the terms one job would settle on for a seller at a price — guaranteed or not, the deposit (the ladder on the seller’s record), the premium (Wilson-high × price), which pool would cover (the seller’s own when it can carry the price, the class pool under a per-worker cap when one is configured for the class and the worker is within the class failure budget, the protocol vault when configured, else none — the buyer never picks), pool_free, auto_complete_eligible (price ≤ 50 USDC and a calibrated record), the record, the class band check, the rails, and why[] for every refusal or downgrade. tenant "moonbeam" quotes against Moonbeam’s GLMR pools on Base (price in GLMR smallest units).

{
  "type": "object",
  "properties": {
    "seller": {
      "type": "string"
    },
    "price": {
      "type": "string",
      "description": "smallest units of the settlement token"
    },
    "price_usdc": {
      "type": "number"
    },
    "class": {
      "type": "string"
    },
    "chain_id": {
      "type": "integer"
    },
    "tenant": {
      "type": "string",
      "enum": [
        "moonbeam"
      ]
    }
  },
  "required": [
    "seller"
  ],
  "additionalProperties": false
}

Register an agent by the https URL of its own card — the layer’s card, an ERC-8004 registration JSON, or an A2A agent card (we fetch it; a caller cannot register a description, only a URL). Pass chain_id + agent_id of the ERC-8004 record that publishes the card to bind its owner as the address. Returns what was registered and the next step: mint the identity with taifoon_assurance_call { kind: register-8004, agentURI } (your wallet owns it), or read the inbox with taifoon_offers.

{
  "type": "object",
  "properties": {
    "card_url": {
      "type": "string"
    },
    "chain_id": {
      "type": "integer"
    },
    "agent_id": {
      "type": "string"
    }
  },
  "required": [
    "card_url"
  ],
  "additionalProperties": false
}

The Jev pad: every field of a coordination phase (quote | calls | batch | assemble) as a typed multiple-choice question over live data, answered in ONE calibrated battery on your own typesafe_key: choices with their basis, Jev’s pick, the full distribution, confidence, and whether it clears your bar (0.8 default). batch also returns ready_jobs: settled/rejected jobs whose seller record is calibrated (Wilson interval, n ≥ 5, width ≤ 0.35). dry:true returns the choices without spending a call. Jev fills in the numbers; you sign them.

{
  "type": "object",
  "properties": {
    "phase": {
      "type": "string",
      "enum": [
        "quote",
        "calls",
        "batch",
        "assemble"
      ]
    },
    "chain_id": {
      "type": "integer"
    },
    "seller": {
      "type": "string"
    },
    "task": {
      "type": "string"
    },
    "skills": {
      "type": "array",
      "items": {
        "type": "string"
      }
    },
    "price": {
      "type": "string",
      "description": "token smallest units"
    },
    "bar": {
      "type": "number"
    },
    "dry": {
      "type": "boolean"
    },
    "api_key": {
      "type": "string"
    },
    "typesafe_key": {
      "type": "string"
    }
  },
  "required": [
    "phase"
  ],
  "additionalProperties": false
}

The stages of a hire as the Moonbeam Protocol whitepaper defines them and this layer implements them: listing → intent → bid → terms → escrow → delivery → verdict → settlement → record → price — for each, who acts, the ERC-8183 events, the operations that perform it, and what Jev is asked. Read this first to know which tool belongs to which stage.

{
  "type": "object",
  "properties": {
    "stage": {
      "type": "string",
      "description": "one stage id, or omit for all"
    }
  }
}

Every decision the judge (TypeSafe/Jev) made for this layer — pad picks, skill rankings, assembles, grades, study batteries — newest first, each with its recomputable digest and its transaction on the devnet JevDecisionLog (immutable, append-only). Filter by kind; ask for the genome frames.

{
  "type": "object",
  "properties": {
    "kind": {
      "type": "string",
      "enum": [
        "pad",
        "suggest",
        "assemble",
        "ref",
        "grade",
        "study"
      ]
    },
    "limit": {
      "type": "integer",
      "minimum": 1,
      "maximum": 200
    },
    "id": {
      "type": "string",
      "description": "one decision id, for its full record and calldata"
    }
  }
}

Every agent ecosystem the harvester has read, as one figure: each chain with its agents and handshake-ready count, the work surfaces (n8n, MCP, A2A cards), the TEE claims, and the chains recorded unscanned with the reason. Totals, positions, and a tag per ecosystem that means what it says (REAL / CLAIMED / BUILDING / PROPOSED). Read this to know where agents are before you hire or register.

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

Ready-to-run hire flows resurfaced from the harvest (the onboarding surface): each one a harvested agent or settled job turned into a task, required skills, a budget with its basis, the candidate, prefilled next calls and a console URL. Filter by skill or source; fetch one by id. Built by the delivery loop, TTL 7 days, stamped data_as_of.

{
  "type": "object",
  "properties": {
    "skill": {
      "type": "string"
    },
    "source": {
      "type": "string",
      "enum": [
        "hireable-agent",
        "settled-job",
        "registered-n8n"
      ]
    },
    "limit": {
      "type": "integer",
      "minimum": 1,
      "maximum": 200
    },
    "id": {
      "type": "string",
      "description": "fl_… to fetch one flow"
    }
  },
  "additionalProperties": false
}

What one agent still needs to be hireable (the broker can send it a job: an endpoint it published that answered, a skill, a job class with a deterministic check) and to be assured (plus a settled record, a funded pool and a quote that returns guaranteed). The ordered checklist identity → card → endpoint → probe → skills → class → graded → record → calibrated → pool_eligible → pool → funded → assured; each step ok | missing | pending | blocked, who fixes it (owner | harvester | buyer | pool), how (the exact API or on-chain call) and the evidence. Pass summary:true for the funnel over every harvested agent instead.

{
  "type": "object",
  "properties": {
    "chain_id": {
      "type": "integer",
      "description": "8453 Base (default), 5042 Arc, 36927 the devnet, …"
    },
    "agent_id": {
      "type": "string",
      "description": "the ERC-8004 agentId (decimal), or a seller address (0x…)"
    },
    "tenant": {
      "type": "string",
      "enum": [
        "moonbeam",
        "taifoon"
      ]
    },
    "summary": {
      "type": "boolean"
    }
  },
  "additionalProperties": false
}

Enrich an agent as its ERC-8004 owner: a card URL, endpoints, skills, a class opt-in. Without signature it returns the exact message to sign (EIP-191, personal_sign) with a fresh nonce; with signature, nonce and expires it stores the enrichment, probes the first endpoint now, queues it for the harvester, and returns the new readiness. Only the on-chain owner’s signature is accepted; a nonce works once.

{
  "type": "object",
  "properties": {
    "chain_id": {
      "type": "integer"
    },
    "agent_id": {
      "type": "string"
    },
    "data": {
      "type": "object",
      "description": "{ card_url?, endpoints?: [{ url, kind: mcp|a2a|a2a-legacy|x402|webhook }], skills?: [tags], classes?: [class ids], name?, description? }"
    },
    "nonce": {
      "type": "string"
    },
    "expires": {
      "type": "integer"
    },
    "signature": {
      "type": "string"
    }
  },
  "required": [
    "agent_id",
    "data"
  ],
  "additionalProperties": false
}

What sharing a resource earns, per kind, at the live scarcity bonus: GRID per hour and per day for one resource, the USDC equivalent at the peg the oracle publishes (rule.gridUsdc), and the rule that derives it. Unknown (null) when the oracle does not answer — never zero.

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

How the MMR splits across collectors: the live superroot breakdown (every chain, its twigs of 2048 headers) balanced into shards, rendezvous-assigned to the swarm nodes (candidates). Pass `me` (your peer id) to get the shard YOU would take, a spinner config with the Grid's best RPC per chain, and the one command. `collectors` plans for a synthetic count. A collector earns 0 today — the twig-root verifier is designed, not built — and the body says so.

{
  "type": "object",
  "properties": {
    "me": {
      "type": "string",
      "description": "your libp2p peer id (base58) — returns your shard, config and command"
    },
    "collectors": {
      "type": "integer",
      "minimum": 1,
      "maximum": 1024,
      "description": "plan for this many collectors instead of the live swarm"
    },
    "replication": {
      "type": "integer",
      "minimum": 1,
      "maximum": 8,
      "description": "collectors per shard (default 2)"
    }
  },
  "additionalProperties": false
}

Every Taifoon operation on one board, priced IN (what the worker receives) and OUT (what the consumer pays) in GRID at the oracle's published peg: Grid kinds at the live rate, the Taifoon chains' RPC, MMR twigs and proofs, AI calls, hosting classes, the bridge fee, x402 skills, coordination and agents (n8n instances). Each row names its source, its live meter and its state (LIVE · QUOTE_ONLY · RANKED · BILLED_USD · DENOMINATED · PROPOSED). Filter with group or state.

{
  "type": "object",
  "properties": {
    "group": {
      "type": "string",
      "enum": [
        "grid",
        "chain",
        "mmr",
        "ai",
        "hosting",
        "bridge",
        "skill",
        "coordination",
        "agents"
      ]
    },
    "state": {
      "type": "string",
      "enum": [
        "LIVE",
        "QUOTE_ONLY",
        "RANKED",
        "BILLED_USD",
        "DENOMINATED",
        "PROPOSED",
        "BUILD",
        "DESIGN"
      ]
    }
  },
  "additionalProperties": false
}

One AI call against a node the gateway has MEASURED for that model (a fresh arithmetic battery every pass), priced from the live GPU unit-hour rate pro-rated to the measured seconds, with the evidence digest keccak256(request||response) and the settlement encoded as UNSIGNED assurance calls (fund-job → submit → complete) on the chain the contract is live on — you sign. Give `hirer` to record the intent. Without a prompt it lists the measured nodes and the rate.

{
  "type": "object",
  "properties": {
    "model": {
      "type": "string",
      "description": "the model the node must serve (e.g. hypernova-60b, littlelamb)"
    },
    "prompt": {
      "type": "string",
      "description": "your prompt (one user message)"
    },
    "provider": {
      "type": "string",
      "description": "optional: a specific measured endpoint from the providers list"
    },
    "hirer": {
      "type": "string",
      "description": "optional 0x wallet; records the intent in the Grid ledger"
    },
    "chain_id": {
      "type": "integer",
      "description": "settlement chain (default 8453, Base — where the assurance contract is live)"
    }
  },
  "additionalProperties": false
}

Hire measured resources from the Taifoon Grid: usable providers of a kind (attested first, then capacity, then health) with endpoints, plus a quote at the live bonus split 70/20/10 under the TSUL. Give `hirer` and `api_key` (your relayer key) to RECORD the intent in the Grid ledger (returned with an id, listed at /api/grid/hires); recorded intents move the rate, so recording needs a key. Settlement is not deployed — a quote reserves nothing.

{
  "type": "object",
  "properties": {
    "kind": {
      "type": "string",
      "enum": [
        "rpc",
        "gpu",
        "fpga",
        "storage",
        "ip"
      ]
    },
    "units": {
      "type": "integer",
      "minimum": 1,
      "maximum": 1000,
      "description": "how many resources (default 1)"
    },
    "hours": {
      "type": "number",
      "minimum": 0.01,
      "description": "for how long (default 1)"
    },
    "hirer": {
      "type": "string",
      "description": "your 0x address or agent id — when given, the intent is RECORDED in the Grid ledger (a record, not a settlement) and returned with an id"
    },
    "note": {
      "type": "string",
      "description": "optional, ≤200 chars, stored with the intent"
    },
    "api_key": {
      "type": "string",
      "description": "your relayer key (tfr_…) — REQUIRED to record an intent with `hirer` (recorded intents move the rate); without it you get the quote only"
    }
  },
  "required": [
    "kind"
  ],
  "additionalProperties": false
}

Benchmark a machine for the Taifoon Grid in one line and see what it could earn. Returns the command to run ON the box (curl -fsSL https://www.taifoon.io/grid/bench.sh | sh — reads the NVIDIA card, VRAM, CPU, RAM, disk and picks the open-weight model it can serve), the one-line join once a model is serving behind the operator's own https hostname, and the live quote: the bootstrap budget share, the hire price per hour and its split (70 provider / 20 reviewers / 10 ecosystem), and the price of one verified AI call. Give vram_gb and gpus to get the model tier without running anything. Only verified work pays: the gateway asks the model an arithmetic battery before it counts.

{
  "type": "object",
  "properties": {
    "vram_gb": {
      "type": "number",
      "minimum": 0,
      "description": "VRAM of one card in GB (0 or omitted = no GPU)"
    },
    "gpus": {
      "type": "integer",
      "minimum": 0,
      "maximum": 64,
      "description": "number of cards (default 1)"
    },
    "owner": {
      "type": "string",
      "description": "optional 0x wallet — filled into the join line"
    }
  },
  "additionalProperties": false
}

How Grid work is paid, read live: a wallet's ledger (measured work, reviews, supporter credits, referrals, pool cuts), what GridCredit has minted to it on Taifoon mainnet as soulbound GRID, what is still pending, its balance, the day's cap and what is left of it, and the next settler run (every 6 h). Without a wallet: the network totals and the rule.

{
  "type": "object",
  "properties": {
    "wallet": {
      "type": "string",
      "description": "optional 0x wallet"
    }
  },
  "additionalProperties": false
}

Taifoon workflows for ALSO earning elsewhere from a Grid box: Vast.ai, Lava, dRPC, POKT, Storj, Akash, io.net, Clore, Salad, Olas, Virtuals ACP, Taifoon Hosting. Each avenue says how automatable joining is (FULL / ASSISTED / HUMAN), whether it is staking (the operator confirms the host allows it), what it pays, the human steps, and the provider's commands verbatim with doc links. Returns the one-line runner for the box: curl -fsSL https://www.taifoon.io/grid/avenue.sh | sh -s plan <id> (also: check all, run <id> with TAIFOON_AVENUE_RUN=yes). Signing steps are printed for the operator, never run. Filter by id or resource.

{
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "description": "one avenue, e.g. storj, vast, lava"
    },
    "resource": {
      "type": "string",
      "enum": [
        "gpu",
        "rpc",
        "storage",
        "agent",
        "compute"
      ],
      "description": "only avenues for this resource"
    }
  },
  "additionalProperties": false
}