ddg-agent-payable-services

DDG Agent-Payable Services (repository name daedalusdevelopmentgroup/ddg-agent-payable-services) is a pay-per-use service gateway for AI agents. Its positioning is to enable AI agents to purchase DDG-packaged inference and bounded outputs without an account, while completing transactions through standardized payment, order, and compliance channels.

In terms of purpose: this project turns Hermes/DDG into a paid service surface open to other AI agents, providing an MCP server and an OpenAI-compatible API for agent tool calls, market data, RPC access, and MCP security auditing. Payment is made in USDC on networks such as Base, with no account registration required.

Main capabilities (based on the current launch wedges in the materials):

• One-cent transaction smoke test (/v1/tx-smoke-test): a receipt-only paid request for other agents to verify that they can complete a DDG paid HTTP transaction, priced at $0.01.
• Site audit (/v1/site-audit): read-only, launch settlement test price $0.75.
• Model routing/basic chat (/v1/model/chat-completions, /v1/model/agent-run): paid access to DDG-operated model routing, supporting basic chat, analysis, code/help prompts, and bounded agent tasks.
• Operator-reviewed OAuth/model artifact runs (service_id=provider_model_artifact_order, via /v1/order-intake): runs DDG-operated account/provider-supported model work only when the terms permit, without reselling raw tokens/accounts/seats.
• Lead pack (/v1/lead-pack): paid local business intelligence, launch price $5.
• Browser proof (/v1/browser-proof): paid screenshot/QA proof, launch price $2.
• Outreach brief (/v1/outreach-brief): paid draft only, not sent without human approval, launch price $1.50.
• Ethereum private RPC queries (/v1/ethereum/rpc): read-only, method-whitelisted Reth proxy, including sync gate, fair-use lease timing, per-agent request cap, maximumbatch and upstream concurrency caps; public self-service requires waiting for execution sync/health.
• Local model free seat pool (/v1/micro-model-swarm-preview, /v1/model/agent-run, /v1/ollama-models): real-time free/best-effort Ollama local routing, one concurrent generation seat per healthy backend, by default at most two for backends configured with 8GB VRAM. Defaults: 24-hour lease per agent, 100 requests per lease, 60-second swarm generation timeout, 120-second local agent-run timeout. The paid lease path is still planned.
• Agent artifact/receipt console (/v1/orders/{order_id}, /v1/orders/{order_id}/artifact): agent-scoped status/artifact URLs returned by /v1/order-intake, exposing redacted receipt/proof hashes and expected deliverables, never exposing raw payment materials.
• Checkout conformance configuration/probe (/.well-known/ddg-agent-checkout-conformance.json, scripts/agent_checkout_conformance_probe.py): lets buyer agents or partner services verify identity gating, 402 challenges, fake-payment fail-closed behavior, idempotency, accepted order links, and artifact access control.
• MPP activation guide (/.well-known/ddg-mpp-activation-guide.md): account creation for operators/buyers, Stripe/Tempo/mppx test setup, required DDG keys, go-live gates, public copy switch checklist, and rollback steps.

How to use:

• Payment protocol: x402 is the real-time HTTP 402 payment flow, aimed at stablecoin-native agents and Bazaar-style discovery. The canonical PAYMENT-REQUIRED accepts[] advertises Base, Polygon, Arbitrum One, World Chain, and Solana mainnet USDC; Base mainnet USDC is listed first for x402scan/CDP compatibility.
• Direct cryptocurrency: real-time receiving address channel, with automatic verification for validator-supported assets, and manual/operator confirmation as a fallback for the rest of the public list. Public families include EVM/stablefixed coins, BTC, BCH, LTC, DOGE, SOL, TRX, XRP, XLM, ALGO, DOT, ZEC, and XMR. ADA/Cardano is intentionally not advertised until there is a DDG-controlled address plus verification/manual confirmation policy.
• MPP: code-ready and signing keys ready, but do not publicly claim launch until real MPP/Tempo currency is configured or Stripe SPT configuration/en, ready:true passes, one-cent payment settlement, idempotent replay passes, fake token fails closed, and leak scan is clean.
• AI-only sales posture: public self-serve and paid routes are limited to AI agents. Human traffic without stable agent identity headers (X-Agent-Id / X-DDG-Agent-Id / X-DDG-User) is rejected with 403 agent_only. What is sold is DDG-wrapped inference and bounded artifacts, not raw model subscriptions/accounts.
• Model router aliases: advertised aliases include auto, mini, ministal, x, y, z, agent-small, agent-standard, agent-pro, agent-max, budget, starter, as well as core DDG aliases glm-4.5-air, glm-4.5, glm-4.6, glm-4.7, glm-5, glm-5-turbo, glm-5.1, glm-5.2, gpt-5.3-codex-spark, gpt-5.3-codex, gpt-5.4, gpt-5.5, claude-haiku-4.5, claude-sonnet-4.6, claude-opus-4.8. Also supports the x_router.agent_profile hint: low / budget / standard / pro / premium.
• Local free seats: the production target is two health-derived 8GB VRAM local runtime seats, exposed only through the payment edge. OLLAMA_API_URLS should only list private generation endpoints in the deployment environment; do not publish raw backend URLs. /v1/ollama-models reports local_free_seats.configured_free_seat_cap, healthy_local_seat_count, and lease cap,does not expose raw LAN URLs. Defaults: 24-hour lease per agent (86400 seconds), 100 requests per lease, 60-second swarm generation timeout, 120-second local /v1/model/agent-run timeout. The same stable agent identity can reuse the free slot after the 24-hour lease expires. Local Ollama routing under /v1/micro-model-swarm-preview and /v1/model/agent-run uses a non-blocking per-backend seat lock, so one busy backend will not block another healthy configured backend. If only one backend is online, the catalog and lease pool are reduced to one seat; if none are healthy, generation returns a controlled unavailable/busy response. The raw local runtime remains private behind the payment edge. Public requests never trigger model pulls or arbitrary model downloads. Operational host/IP/GPU inventories belong only to private operations notes and do not appear in the startup documentation consumed by agents.

Use cases: for AI agents that need pay-per-call access to external tools, market data, blockchain RPC, security audits, website audits, screenshot proof, outreach drafts, lead intelligence, and other services, as well as agent workflows that need accountless, stablecoin-settled access to LLM inference and bounded tasks; especially suitable for agent CI/smoke tests that want to verify they can complete the x402 paid transaction flow.

Limitations and caveats:

• Public self-serve and paid routes are limited to AI agents; human traffic is rejected with 403.
• Do not resell raw API keys, auth tokens, or account seats; do not advertise as unrestricted resale of OpenAI/GLM/Claude accounts.
• Claude Pro/Max subscription access should not be exposed to buyers; Claude aliases are only for API-supported DDG application output.
• GPT-5.3 Codex Spark/Codex naming and public API availability need to be verified before broad sales; OpenAI Codex documentation says Spark is a research preview/Pro-only at release.
• Ethereum RPC public self-serve should wait for Reth execution sync, non-optimistic consensus state, and DDG_RETH_RPC_URL to be healthy via the payment edge.
• When production mode is enabled but no verifier URL is configured, paid routes return 503 payment_verification_unavailable.
• If no direct cryptocurrency verifier URL, the agent needs to submit payment_proof at /v1/order-intake for confirmation by the operator.
• The materials do not provide specific price list details, SLA, support channels, etc., so they are not provided for now.