Configure a station
A station is a Convoy deployment: a tenant's APIs, plus one configuration file. It opens through two doors:
- other people's agents — Claude, ChatGPT, Cursor — through MCP;
- our agent — the widget on the tenant's site, and soon Teams, Slack or an API.
Both doors read the same config and apply the same rules. What a person does on one side shows up on the other, in their room.
Two kinds of stations
| connected to | example | |
|---|---|---|
| Product | the APIs of the product being made agentic | a restaurant, a shop, a legal service |
| Process | the APIs of several internal tools | a support ticket that ends in a refund |
Both are written with the same primitives: skills, a selection, a submit, playbooks. What changes is the config, not the code.
Two versions of this guide
The menu at the top right switches version.
- Current (this version) — what Core loads today. Every example in this version is compiled by Core's test suite: if it no longer loads, the build fails. What you read here works.
- Target · roadmap — where we will be at the end of the roadmap: profiles, policies, intent-based activation, automations, approvals. It follows ADR-028 §6, ADR-029 and ADR-030. Its examples do not load yet, and that is expected.
The Current limits page says what is missing today, and where to find it in the target version.
How to read this guide
- Concepts — the vocabulary, on one page.
- Configure, step by step — one page per block of the config, in the order you write it.
- Examples — three levels (simple, medium, advanced), for a product station and for a process station. Each example shows the full file, then what happens when you use it.