AI agents pay AI agents.

Five-line Python — agent pays 0.10 USDC, gets the latest market gaps for an AI/ML sector:

# pip install solana solders requests
from agent_demo import x402_get  # see examples/agent_demo.py

resp = x402_get("https://gapsmith.draftlabs.org/api/v1/scout/gaps?sector=ai-ml",
                wallet_secret_key=YOUR_PRIVATE_KEY_BYTES)
print(resp.json())  # → { gaps: [...], count: N, generatedFrom: "2026-..." }

Reference implementation: agent_demo.py — self-contained, ~150 lines. Request a copy at /contact.

All /api/v1/* endpoints run on a cost-effective LLM tier(MiniMax M2.7 / Claude Sonnet 4.6 class). That's what makes per-call pricing in the $0.05–$15 range sustainable — we're not running every probe on Opus 4.7.

Need top-tier output? The Done-For-You service runs the same pipelines on Claude Opus 4.7 / GPT-5.5 Pro with human review. Slower and pricier ($39 / $99 / $149 per run, 24-72 hours), but built for the moments where you ship the report to a customer or investor.

Three round trips — but the second is on-chain, not HTTP:

1. Probe: Agent makes the request without payment. Server returns HTTP 402 Payment Required with a JSON body listing acceptable payment options (recipient, amount, mint, network).
2. Settle: Agent constructs an SPL transferChecked instruction sending the required USDC to our merchant ATA, signs with its own Solana wallet, and submits to the network. Wait for confirmation (~2s).
3. Redeem: Agent retries the same request with X-PAYMENT header containing the transaction signature. Server verifies on-chain, then returns the resource.

402 response body

HTTP/1.1 402 Payment Required
Content-Type: application/json
x402-version: 1

{
  "x402Version": 1,
  "accepts": [{
    "scheme": "exact",
    "network": "solana",
    "maxAmountRequired": "100000",        // atomic units (USDC has 6 decimals)
    "asset": "EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v",
    "payTo": "DtESM4kaLx6Kjxz5oTKwv4AxkkEbsphHcMJXV4ShRrZ5",
    "resource": "https://gapsmith.draftlabs.org/api/v1/scout/gaps?sector=ai-ml",
    "description": "Scout market gap intelligence — aggregated from recent multi-agent scans",
    "mimeType": "application/json",
    "maxTimeoutSeconds": 60
  }],
  "error": "Payment required..."
}

X-PAYMENT header format

Base64-encoded JSON. Raw JSON also accepted (for ergonomic curl testing).

{
  "x402Version": 1,
  "scheme": "exact",
  "network": "solana",      // or "solana-devnet"
  "payload": {
    "txSignature": "<base58-encoded tx signature>"
  }
}

Devnet for testing

Append ?network=devnet to any endpoint URL to negotiate payment in devnet USDC instead of mainnet. The 402 response will include the devnet USDC mint, devnet merchant ATA, and the X-PAYMENT must reference a devnet tx. Mainnet is the default (no param needed) for production agent traffic.

# Test for free with devnet USDC (faucet.circle.com)
curl -i "https://gapsmith.draftlabs.org/api/v1/scout/gaps?sector=ai-ml&network=devnet"

# Switch to mainnet for real agents (default)
curl -i "https://gapsmith.draftlabs.org/api/v1/scout/gaps?sector=ai-ml"

Idempotency

• Same tx_hash on the same endpoint → returns cached response (sync) or 409 (async)
• Same tx_hash on a different endpoint → 400 rejection (proof was for another resource)
• Wrong amount → 400 amount mismatch
• Wrong mint (e.g. USDT instead of USDC) → 400 mint mismatch

Production calls to /api/v1/* default to Solana mainnet — settlement is real USDC. For integration testing without burning funds, append ?network=devnet to the request URL. The server then advertises devnet USDC mint + a devnet merchant ATA in its 402 response, and verifies the signed tx on Solana devnet.

# Probe → 402 advertising devnet USDC mint + devnet merchant ATA
curl -i 'https://gapsmith.draftlabs.org/api/v1/scout/gaps?sector=ai-ml&network=devnet'

# 402 body (abridged):
# {
#   "accepts": [{
#     "network": "solana-devnet",
#     "asset": "4zMMC9srt5Ri5X14GAgXhaHii3GnPAEERYPJgZJDncDU",   ← devnet USDC mint
#     "payTo": "<devnet merchant ATA>",
#     "maxAmountRequired": "100000"   // 0.10 USDC, atomic
#   }]
# }

# Fund a devnet wallet:
#   SOL:  https://faucet.solana.com (request 1 SOL)
#   USDC: https://faucet.circle.com (request a few USDC on Solana devnet)
# then sign + submit a transferChecked tx to the advertised ATA, retry with X-PAYMENT.

examples/agent_demo.py runs on devnet by default — pass --mainnetwhen you're ready to settle in real USDC. The exact same code path runs in both modes; only the network query parameter, mint, and merchant ATA differ.

Data API · synchronous · cache-only

Pre-aggregated market intelligence from recent Scout runs. Returns in <1 second. Ideal for AI agents querying market data at scale.

All accept ?sector= filter and ?limit= cap. /trends also takes ?days=.

Discovery — call free first

Data API responses are filtered from cached Scout runs. If you query ?sector=fintech but only ai-ml + healthtechare cached, you'll pay USDC and get an empty array. Hit /api/v1/sectorsfirst (free) to see what's available.

Compute API · async · runs full pipeline

Triggers a real multi-agent pipeline. Returns 202 Accepted + jobId immediately; agent polls /api/v1/jobs/{id} until completion or supplies a webhook_url for push delivery.

Job management · free

Free because jobId already encodes the capability — knowing the id implies you paid for it.

Compute API endpoints can take 10-45 minutes. We don't hold an HTTP connection that long — instead:

# 1. Pay + dispatch
resp = x402_post("https://.../api/v1/forge/ideate", body={
    "sectors": ["ai-ml"],
    "context": "Find SaaS gaps in agent observability tooling",
    # Optional: calibrate to your real constraints. Pass either a structured
    # object (preferred — enums caught by zod) or raw SESSION_CONFIG.md.
    # Without it the engine falls back to default Small Team / $10K / 4-8 weeks / $100K.
    "session_config": {
        "profile": "Solo",                # Solo | Small Team (2-3) | Small Team (4-5) | Funded Team (6-15) | Enterprise
        "budget": "$1K",                  # $1K | $5K | $10K | $25K | $50K | $100K+
        "timeline": "3-6 months",         # 2 weeks | 4 weeks | 4-8 weeks | 8-12 weeks | 3-6 months
        "revenue_threshold": "$50K/year", # $10K/year | $50K/year | $100K/year | $500K/year | $1M+/year
        "founder_signal": "8 years payments infra, ex-Stripe."
    },
    "webhook_url": "https://my-agent.com/forge-callback"   # optional
})
# resp.status_code == 202
job = resp.json()
# { "jobId": "fg_abc123", "statusUrl": "/api/v1/jobs/fg_abc123", "etaMinutes": 35 }

# 2. Poll (free, no payment needed)
import time
while True:
    s = requests.get(f"https://gapsmith.draftlabs.org/api/v1/jobs/{job['jobId']}").json()
    if s["status"] == "completed":
        print(s["result"])
        break
    if s["status"] == "failed":
        raise RuntimeError(s["error"])
    time.sleep(60)

# 3. (Optional) Webhook delivery
# If you set webhook_url, we POST to it with HMAC-signed body.
# Header: x-gapsmith-signature: <hex(hmac_sha256(secret, body))>
# Body: { "jobId": "fg_abc123", "status": "completed", "result": {...} }

Job statuses

• Stable pricing — USDC is 1:1 with USD, no FX or volatility. 0.10 USDC always means $0.10.
• Sub-cent settlement fees— Solana network fees are ~0.000005 SOL (<$0.001 at typical SOL prices). The whole API economics rely on Solana's low fees being lower than the smallest amount worth charging.
• Standard HTTP semantics — Nothing custom. Any HTTP client, any language, can call this API. No SDK lock-in.
• No chargebacks— On-chain settlement is final. Agents can spend confidently, sellers don't deal with disputes.
• Designed for autonomous agents— No signup, no account, no API key rotation. The agent's own wallet IS its identity. Coinbase and Anthropic explicitly designed x402 for this use case.