> ## Documentation Index > Fetch the complete documentation index at: https://docs.stateset.com/llms.txt > Use this file to discover all available pages before exploring further. # Launch a brand that runs itself > From nothing to a brand operating autonomously on the StateSet iCommerce Engine — onboard across six services, connect the commerce stack, put agents on every channel, gate what moves money, and close the loop on outcomes. The capstone: every step links to a verified walkthrough. This is the whole thesis in one path: **commerce that runs itself**. Each phase below is one sitting, each links to a guide that has been verified against the running service, and by the end a brand is answering customers, executing the operational fixes behind those conversations, and proving every consequential decision — with humans approving only what crosses the lines you set. ```mermaid theme={null} flowchart LR A[1 · Onboard] --> B[2 · Connect commerce] B --> C[3 · Agents on every channel] C --> D[4 · Gates and workflows] D --> E[5 · Measure outcomes] E --> F[6 · Improve the agents] F -.retrained model.-> C ``` [onboarding.stateset.com](https://onboarding.stateset.com) provisions everything from a plain-language intake: the org, a knowledge base with your policies embedded, a ResponseCX organization with support rules, the brand in the Temporal workflow engine, NSR's `ecommerce-returns` rulebook with your refund window and final-sale rules, and an unpublished Voice agent. Preview before launch; launch is idempotent; verification probes all six services and heals drift. → [Brand Onboarding](/stateset-onboarding) — portal, operator CLI, provisioning API, and the MCP server an agent can drive itself. Two commands anchor the whole phase — provision, then prove it: ```bash theme={null} onboard-brand acme-outdoors # one line per service as it provisions onboard-brand verify acme-outdoors # present / absent / partial, per service; exit 0 = fully onboarded ``` Onboarding **never invents your business rules** — what you leave blank stays blank. Write your shipping, returns, escalation and tone policies as plain language; the translator turns them into KB articles, ResponseCX rules and NSR clauses deterministically. Point the Sync Server at Shopify and NetSuite (or your ERP/3PL): orders flow in, inventory flows back, every job is idempotent and replayable, webhooks verified. If retailers send you purchase orders, EDI turns their 850s into orders and your 856s/810s into their inbox. → [Sync a Shopify store with NetSuite](/guides/sync-shopify-to-netsuite) · [Your first EDI 850 cycle](/guides/edi-first-850) · [A retailer's 850 to cash](/guides/journey-edi-order-to-cash) The ResponseCX agent onboarding created is inactive until you launch it — attach knowledge, give it functions (real API calls, tested before a customer finds out), constrain it with rules, then activate. Embed the chat widget on the storefront, publish the Voice agent once you have heard it in your browser, and hand the back-office tools that have no API to Computer Use. → [ResponseCX quickstart](/quickstart) · [Chat Widget](/stateset-response/response-chat-widget) · [Put an agent on the phone](/guides/voice-first-phone-agent) · [Automate a tool that has no API](/guides/computer-use-first-task) Autonomy without control is a liability. Refunds, credits and cancellations route through NSR — `approved`, `denied`, or `refused` **with the cited rules**, fail-closed when the policy engine is unreachable, human approval above your value threshold. The operational work itself runs as Temporal workflows: retried, compensated, replayable, with the proof chain in the permanent event history. The gate is one call — and a `refused` verdict is a **200 with reasons**, not an error: ```bash theme={null} curl --request POST https://api.nsr.stateset.com/v1/decisions \ --header "X-API-Key: $STATESET_NSR_API_KEY" \ --header "Content-Type: application/json" \ --header "Idempotency-Key: refund-ORD-10042" \ --data '{ "authorization_goal": "authorize_refund", "external_ref": "ORD-10042", "facts": [ { "predicate": "refund_requested", "args": ["ORD-10042", "84.00"] } ] }' ``` → [Gate a refund with a verified decision](/guides/nsr-first-verified-decision) · [Run a durable order workflow](/guides/temporal-first-durable-workflow) · [ResponseCX approvals](/stateset-response/responsecx-autonomy) A "where is my order?" message should end with the order actually fixed — this journey follows one message across ResponseCX, Sync, Temporal and NSR to a closed operation. Billing then charges for **verified outcomes**, not seats: a resolved contact is billable work, a disputed one is a credit. → [From "where is my order?" to a closed operation](/guides/journey-wismo-to-resolution) · [Billing and outcomes](/stateset-billing) StateSet Agents grades production conversations against a reward, curates what worked, retrains, and evaluates on prompts the model never saw — lineage from the deployed model back to the conversations that produced it. This is the difference between an agent that changed and an agent that got better. → [StateSet Agents](/stateset-agents/overview) · [Getting started for AI agents](/getting-started-for-ai-agents) ## What "autonomous" means here, concretely | The brand does | Autonomously? | The control | | ---------------------------------------- | ---------------------------- | ------------------------------------------------------------------ | | Answer customers on chat, email, phone | Yes | Rules, knowledge, evals; escalation phrases hand off to a human | | Look up orders, shipments, subscriptions | Yes | Read-only functions, tested before launch | | Fix the operation behind the ticket | Yes, within policy | Temporal workflows with retries, compensation, replay | | Refund, credit, cancel | Only with a verified verdict | NSR cited-proof gate, fail-closed, human approval ≥ your threshold | | Operate tools with no API | Yes, budgeted | Computer Use cost caps, approval policies, full replay | | Get better over time | Yes, provably | Eval suites and held-out prompts before a retrained model ships | Run this whole page as an agent instead of a person: every service here has an [MCP server](/mcp-servers), the skills are one `curl` away, and [onboarding itself has MCP tools](/stateset-onboarding) — an agent can provision the brand, connect the stack, and launch the channels. That is the point. ## When something doesn't fit Each linked guide carries its own troubleshooting; [Getting Help](/support) routes everything else — host status, the cross-engine error taxonomy, Discord, and the outcome dispute process. ## Next steps The onboarding portal is live — intake to verified provisioning in a sitting. One customer message crossing four engines to a closed operation. MCP servers and skills for every phase — including onboarding itself. Hosts, keys, headers and first calls for all twelve surfaces.