§ docs · four integrations, one spine
Human quickstart on top.
Agent manifest layer, untouched.
The existing agent-facing content (Discovery, Agent Guides, API specs, LLM Index) is architecturally correct for discovery. We added a human lane on top without disturbing the machine-readable layer below.
§ pick a path
Four surfaces. Same operation.
Same store-and-verify operation, four different surfaces. The one you pick is a token-efficiency choice first, an ergonomics choice second. Context-efficiency data is under /proof/context-efficiency.
HTTP API
~1,380 tok/opRaw gateway — any language, any runtime. Envelope construction, hash verification, merkle walk in-prompt.
PyNukez SDK
~510 tok/opPython-native. Typed client, async, full lifecycle in a dozen lines. Framework adapters for LangChain and LlamaIndex.
MCP server
~904 tok/opClient-signed envelopes. 14 tools exposed, matching the tool_contract.json schema. Works with any MCP-speaking client.
NukezAgent (hosted)
~162 tok/opNatural language to receipt. Zero protocol in your prompt. $500/mo flat. Four framework adapters out of the box.
§ agent surface · machine manifests
Fetch these first if you're an agent.
Discovery manifests
- GET /.well-known/nukez.json — top-level manifest
- GET /.well-known/nukez-agent.json — A2A discovery
- GET /v1/tools.json — MCP tool schema (14 tools)
- GET /v1/tool_contract.json — structured contract
- GET /openapi.json — OpenAPI spec
- GET /llms.txt — LLM index
§ operations
Portal and passkey setup are operations guides — they belong next to each integration, not as top-level marketing. See /docs/portal for viewer + owner portals and passkey setup; see /docs/verify for the dev-facing verification reference.