Agent-Native Commerce: Introducing the PlatformDTC Agent Gateway
Chatbots talk. Agents act. One secure, programmable API that lets autonomous agents build stores, import catalogs, publish to a live domain, launch campaigns and read back what they earned.

Chatbots Talk. Agents Act.
For two years, AI in commerce mostly meant a chat box that answers questions. Helpful — but it does not run the store.
Today PlatformDTC is shipping the piece that does: the Agent Gateway, one secure programmable API that lets autonomous agents take real actions across an entire store.
What It Is
A versioned API spanning 36 endpoints across store building, catalog, advertising, marketing, analytics and orders — mirrored one-to-one by 23 tools over the Model Context Protocol, so an agent registers a single server and gets the whole platform.
An agent should be able to build a storefront from a brief, source and import products, publish to a live domain, connect ad accounts, launch a campaign, and watch return on ad spend — then decide what to do next. Doing that safely against real money and real customers needs more than a model with a prompt. It needs a control plane.
One Sequence: Brief to Store to Ads to Return
The headline flow an agent runs is a single unbroken sequence. Create a store from a brief. Bulk import a catalog. Publish the project to a production domain. Connect an ad platform, which hands off to a human to authorize. Create and send a campaign, which may park for approval. Read the dashboard back and see revenue, orders, ad spend and return.
Every long step returns a job the agent waits on, then moves past. Every spend step can require a person.
The same calls work over plain HTTP with an API key. Anything an agent can do is reproducible from a terminal, which means the platform's behaviour is auditable by hand rather than only observable through an agent.
Built for Real Operations
- Least privilege — per-agent keys carry only the scopes granted, and spend actions are gated
- Safe retries — idempotency keys are mandatory, so duplicate calls never double-charge
- Async by design — long operations return a job, with one surface to poll for every workflow
- Full audit — every call recorded with principal, tool, scopes, status and latency
- Signed webhooks — HMAC-signed event callbacks so agents react instead of polling
- Human approval — an optional gate on spend, where agents propose and humans confirm
Why This Is the Company's Direction
PSA's stated goal is an autonomous commerce system that can generate profit by itself. A gateway that lets an agent act, under permissions a human sets, is the mechanism that makes that goal a product rather than a position.
The launch write-up is at platformdtc.com/blog, and the security model that contains it is published alongside.