Concierge Chat

POST /api/concierge · 0.1 USDC (or equivalent in a live Token Pay merchant token on Solana) · Gemini 2.5 Flash (default) or optional GLM-4.7 Flash (Z.ai) with live market, DeFi desk, lounge memory, and trading-plan structure.

• USDC — Base or Solana via PayAI/Dexter (standard x402 client)
• Native SPL — Token Pay self-settle when your merchant is registered and live (GET /api/token-pay?merchant=YOUR_ID)

402 responses may include multiple accepts. Pick token pay when extra.settlement === "self" and extra.merchantId matches your registry row.

{
  "mode": "chat",
  "message": "Hottest Meteora DLMM pools right now — IL risks?",
  "history": [
    { "role": "user", "text": "…" },
    { "role": "model", "text": "…" }
  ],
  "market": [],
  "agentModel": "gemini"
}

Optional agentModel: gemini (default) or glm-4.7-flash (Z.ai — server needs GLM_API_KEY). Image, enhance, and trading-plan paths always use Gemini. GLM failures fall back to Gemini (modelFallback: true).

Optional header: X-Agent-Id: agt_… after registration.

{
  "reply": "<p>…</p>",
  "topics": ["crypto", "defi"],
  "marketLive": [{ "symbol": "BTC", "price": "…", "change": "…" }],
  "dataAsOf": "2026-05-21T12:00:00.000Z",
  "modelUsed": "gemini-2.5-flash",
  "images": ["data:image/png;base64,…"]
}

Yield / DLMM questions use live Meteora DLMM API + DeFi Llama in the prompt. DLLM typos are interpreted as DLMM.

Trading-plan questions may include A2A|asset=…|bias=… for downstream agents.

npm install @x402/core @x402/fetch @x402/evm viem

Wrap fetch with wrapFetchWithPayment and registerExactEvmScheme on Base — full example in Quickstart and PayAI reference.

• Message max 4,000 characters · body max 48 KB
• History max 12 turns
• Edge timeout ~30s — trading-plan path is optimized; Intel APIs are faster for structured JSON