SchemAgentic
Documentation
SchemAgentic is a canvas of nodes, and a node can be more than something you wrote down: it can have an agent that keeps working, and services that let it reach the tools you already use. These two pages are the whole of how that works.
Agents
Something that runs on a node, on a schedule or when you ask, after you have closed the document.
- What an agent can see and do
- When it runs, and how often
- What it must ask you about first
- Limits, plans and what it costs
Connections
What travels from one node to another, and when: ports, push, pull and trigger.
- The three modes
- Filters and reshaping
- What an agent can send and read
- Limits, and why a circle stops
Services
How a node reaches the outside world — and the only way an agent can reach anything at all.
- The built-in catalogue
- Credentials, and why you never see one again
- Payments, limits and approvals
- Every rule a call has to pass
Chat
Two conversations, and which one you are in: building the schema, and talking to an agent about its own work.
- Building by describing what you want
- Adding services and agents by asking
- Why a credential never goes through the chat
- Talking to one agent, and why it proposes rather than acts
AI
Where the AI actually is, what a turn costs, and what leaves your machine.
- The four places AI appears
- Choosing a model
- Allowances, and who pays in a shared schema
- What is sent, and what never is
MCP server
The other direction: an AI agent elsewhere reading and adding to your maps.
- Publishing maps, and generating a token
- Connecting Claude Code or Claude Desktop
- The five tools
- Why a write is queued, not applied