DDG Agent Services

mcp-name: com.daedalusdevelopmentgroup/ddg-agent-services-mcp

DDG Agent-Payable Revenue Launch Pack

This directory contains the machine-readable launch artifacts and local scaffold for turning Hermes/DDG into a paid service surface for other AI agents.

Current launch wedges

1. One-cent transaction smoke test (/v1/tx-smoke-test) — $0.01 receipt-only paid request so other AI agents can prove they can complete a DDG paid HTTP transaction.
2. Website audit (/v1/site-audit) — read-only; launch settlement-test price $0.75.
3. Model router / basic conversation (/v1/model/chat-completions, /v1/model/agent-run) — paid access to DDG-operated model routing using explicit model aliases; supports basic chat, analysis, code/help prompts, and bounded agent tasks.
4. Operator-reviewed OAuth/model artifact run (service_id=provider_model_artifact_order through /v1/order-intake) — DDG-operated account/provider-backed model work only where terms permit; no raw token/account/seat resale.
5. Lead pack (/v1/lead-pack) — paid local business intelligence; launch price $5.
6. Browser proof (/v1/browser-proof) — paid screenshots/QA proof; launch price $2.
7. Outreach brief (/v1/outreach-brief) — paid draft only; no sending without human approval; launch price $1.50.
8. Ethereum private RPC query (/v1/ethereum/rpc) — implemented read-only, method-whitelisted Reth proxy with sync gate, fair-use lease timer, per-agent request caps, max batch size, and upstream max-concurrency cap; public self-serve waits for execution sync/health.
9. Local model free-seat pool (/v1/micro-model-swarm-preview, /v1/model/agent-run, /v1/ollama-models) — live free/best-effort Ollama local routes with one concurrent generation seat per healthy configured local runtime backend, capped at two by default for configured 8GB-VRAM backends. Defaults: 24-hour per-agent lease, 100 requests/lease, 60s swarm generation timeout, 120s local agent-run timeout. The named paid lease path remains planned.
10. Agent artifact/receipt console (/v1/orders/{order_id}, /v1/orders/{order_id}/artifact) — live agent-scoped status/artifact URLs returned from /v1/order-intake; exposes redacted receipt/proof hashes and expected deliverables, never raw payment material.
11. Checkout conformance profile/probe (/.well-known/ddg-agent-checkout-conformance.json, scripts/agent_checkout_conformance_probe.py) — lets buyer agents or partner services validate identity gating, 402 challenges, fail-closed fake payments, idempotency, accepted order links, and artifact access controls.
12. MPP activation guide (/.well-known/ddg-mpp-activation-guide.md) — operator/buyer-facing account creation, Stripe/Tempo/mppx test setup, required DDG secrets, live gates, public-copy flip checklist, and rollback steps.

Penny transaction smoke test

POST /v1/tx-smoke-test is the tiny paid-request endpoint. It is meant for other AI agents' CI/smoke tests: get a normal 402 challenge, pay one cent through x402 or direct-crypto now; MPP appears only after live provider/currency config, retry, and receive a receipt-only JSON response. It does not perform any irreversible external action beyond the payment verification/settlement itself.

Ethereum/Reth privacy-node posture

The /v1/ethereum/rpc route is for accountless buyer-side Ethereum reads: DDG runs the Reth node/proxy, the buyer agent does not need an Infura/Alchemy/vendor account, and the edge still requires AI-agent identity. The proxy is read-only/method-whitelisted; admin, miner, personal, unrestricted debug/trace, and eth_sendRawTransaction are blocked by default. The free beta wrapper now includes:

• sync readiness gate: DDG_ETH_RPC_REQUIRE_SYNCED=1 returns 503 reth_rpc_syncing until eth_syncing=false and latest block is non-zero/current;
• fair-use timer: DDG_ETH_RPC_FREE_MAX_CONCURRENT_USERS=4, DDG_ETH_RPC_FREE_LEASE_SECONDS=600, DDG_ETH_RPC_FREE_MAX_REQUESTS_PER_LEASE=30 by default;
• node overload guard: DDG_ETH_RPC_MAX_CONCURRENT=2, DDG_ETH_RPC_MAX_BATCH=10, bounded timeout/response bytes.

Public self-serve should wait for Reth execution sync, consensus non-optimistic status, and DDG_RETH_RPC_URL health through the payment edge.

Protocol stance

• x402: live HTTP 402 payment flow for 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 stays first for x402scan/CDP compatibility.
• Direct crypto: live receiving-address rail with automatic verification for verifier-supported assets and manual/operator-confirmed fallback for the rest of the public manifest. The public families are EVM/stablecoins, BTC, BCH, LTC, DOGE, SOL, TRX, XRP, XLM, ALGO, DOT, ZEC, and XMR. ADA/Cardano is intentionally not advertised until a DDG-controlled address plus verification/manual-confirmation policy exists.
• MPP: code-ready and signer-key-ready, but not publicly claimed live until a real MPP/Tempo currency or Stripe SPT profile/env is configured, ready:true passes, a penny payment settles, idempotent replay passes, fake tokens fail closed, and the leak scan remains clean.

AI-only sales posture

• Public self-serve and paid routes are AI-agent only.
• Human traffic without a stable agent identity header (X-Agent-Id / X-DDG-Agent-Id / X-DDG-User) is rejected with 403 agent_only.
• We sell DDG-packaged inference and bounded artifacts, not raw model subscriptions/accounts.

Model-router model aliases

Advertised aliases: auto, mini, ministal, x, y, z, agent-small, agent-standard, agent-pro, agent-max, budget, starter, plus 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.

x_router.agent_profile hints are also supported for non-advanced callers: low/budget/standard/pro/premium.

Corrections from research:

• Z.AI docs list GLM-4.5-Air; I did not find an official GLM-4.6-Air page/pricing row.
• Anthropic docs list Claude Opus 4.8, Sonnet 4.6, and Haiku 4.5; I did not find official Sonnet 4.7 in the current docs.
• Claude Pro/Max subscription access should not be exposed to buyers. Claude aliases are for API-backed DDG application output only.
• GPT-5.3 Codex Spark/Codex naming and public API availability must be verified before broad sale; OpenAI Codex docs describe Spark as research-preview/Pro-only at launch.

Model-router compliance stance

Do not sell raw API keys, auth tokens, or account seats. Do not market this as unrestricted resale of OpenAI/GLM/Claude accounts. The safer packaging is:

Agents buy DDG-operated inference/agent-task outputs with budgets, receipts, abuse controls, and model-tier routing.

Local inference/free-seat status — 2026-06-14

The live implementation can expose up to two free local-model seats when configured local runtime backends are healthy:

• Production target: two health-derived 8GB-VRAM local runtime seats, exposed only through the payment edge.
• OLLAMA_API_URLS should list private generation endpoints in deployment env only; do not publish raw backend URLs.
• /v1/ollama-models reports local_free_seats.configured_free_seat_cap, healthy_local_seat_count, and lease caps without exposing raw LAN URLs.
• Defaults are intentionally usable but bounded: 24-hour per-agent lease (86400 seconds), 100 requests per lease, 60s swarm generation timeout, and 120s local /v1/model/agent-run timeout. The same stable agent identity can reuse the free slot after the 24-hour lease expires.
• /v1/micro-model-swarm-preview and local Ollama routes under /v1/model/agent-run use non-blocking per-backend seat locks, so one busy backend does not block another healthy configured backend.
• If only one backend is online, the catalog and lease pool drop to one seat; if none are healthy, generation returns a controlled unavailable/busy response.
• Raw local runtimes stay private behind the payment edge. Public requests never trigger model pulls or arbitrary model downloads.

Operational host/IP/GPU inventory belongs in private ops notes only, not in agent-consumed launch docs.

Files

• pricing.json — canonical draft price catalog.
• agent-catalog.json — .well-known AI-agent catalog that summarizes discovery URLs, live routes, manual review offers, required headers, and safety posture.
• quickstart.md — concise agent-client onboarding/checklist for 403/402/idempotency/payment behavior, including recommended first order-intake packages and examples.
• sales-launch-pack-2026-06-08.md — customer-facing beta launch package sheet with first services to sell, copy, examples, and current caveats.
• reth-ethereum-agent-rpc-analysis-2026-06-08.md — decision note for planned Reth/Ethereum node access as an agent-facing blockchain intelligence/RPC product.
• openapi.agent-services.json — OpenAPI 3.1 discovery with MPP-style payment metadata.
• agent-checkout-conformance-profile.json — machine-readable 403/402/order-console/idempotency conformance profile exposed at /.well-known/ddg-agent-checkout-conformance.json.
• mpp-tempo-stripe-account-activation-guide-2026-06.md — step-by-step MPP/Tempo/Stripe account, env, smoke-test, go-live, and rollback guide exposed at /.well-known/ddg-mpp-activation-guide.md.
• llms.txt — agent-readable service catalog.
• edge/payment_edge_scaffold.py — payment edge with pluggable verification adapters. Production uses provider-backed verification; mock tokens are development-only and must remain disabled in systemd production env.
• scripts/operator_queue_digest.py — safe operator digest builder for queued paid-agent orders; writes Telegram-ready markdown but does not send or fulfill anything automatically.
• scripts/agent_checkout_conformance_probe.py — non-spending by default checkout probe; with dev mock/real payment credentials it validates accepted order links, artifact/status access controls, and idempotency behavior.
• edge/agent_runners.py — sandbox-conscious subprocess wrappers for env-gated Kimi/Claude operated artifact routes.
• mcp/README.md and mcp/ddg-agent-swarm-mcp-design.md — MCP quickstart/design for AI-agent swarms that want service discovery, quotes, paid model runs, order intake, security-service catalog access, and one-cent transaction smoke tests.
• edge/mpp_verifier_sidecar.mjs — seller-side MPP verifier sidecar using mppx/server; requires recipient, currency, and a Tempo server signing account for real settlement.
• edge/x402_verifier_sidecar.py — x402 verifier/settle adapter translating the Python edge's generic payment payload into facilitator paymentPayload / paymentRequirements calls.
• edge/direct_crypto_verifier_sidecar.py — public-chain direct-crypto verifier sidecar for BTC/LTC/DOGE/BCH/ZEC, EVM native transfers, and USDC on Ethereum/Base/Polygon; unsupported chains fail closed to manual confirmation.
• payment-provider-and-waf-runbook-2026-06-05.md — setup walkthrough for MPP/x402 verifier URLs, direct crypto/manual fallback, and free Cloudflare 1010 alternatives.
• deploy/two-host-local-ai-production-runbook.md — private deployment checklist for the two free local AI seats, lease policy, runtime status, and post-sync verification.
• direct_crypto_addresses.public.json / direct_crypto_addresses.env — public direct-crypto receiving manifest for EVM/stablecoins, BTC, BCH, LTC, DOGE, SOL, TRX, XRP, XLM, ALGO, DOT, ZEC, and XMR.
• audit/preservation_snapshot.md — branch/status snapshot taken before launch edits.

Local scaffold smoke test

python3 sales_artifacts/agent_payments/edge/payment_edge_scaff