Agent Control Plane — Studio Web Guide

agent-control-plane is the open-source control plane for sideseat. Agents reach the stack through WORKER_BASE_URL and the tool registry—not the DB directly—and Studio is where you edit Object·Link·Action graphs and workflows. Paths and env follow the repo README.

Repository & runtime stack

Studio ↔ Ontology ↔ Worker (overview)

Products & workflows Runtime outcomes and automation—contracts you define in Studio flow here

★ Studio — platform-web Node editor · Object / Link / Action · workflow view · Save / Execute / Demo / export

Worker / API Think → Act → Observe · tools · policy

Ontology contract Schema, links, allowed actions—pinned as a graph

Docker roles ops / backend / support boundaries

Governance Only allowed actions via Worker/registry—same idea as “no direct DB”

How orchestrators, sub-agents, and skill agents read the ontology matches the same mental model as long-form “architecture guide” layouts. Here, Studio is the tool that pins that contract in the UI.

agent-control-plane — GitHub

https://github.com/calicorone/agent-control-plane

Vite+React Studio · Docker roles (ops/backend/support)

Part 2

Studio view

A typical visual IDE layout.

Left: sideseat CONTROL PLANE navigation
Center: dot-grid canvas
Right: per-node properties panel

The top bar lines up OPEN, NAME, Save / Execute / Demo / Delete, export (.export.json, .workflow.xml), and the API Bearer token field.

sideseat CONTROL PLANE — Studio: ontology and workflow graph

Local dev. Object·Link·Action and workflow on one canvas.

The center graph is a Meetup-creation example.

Purple Object nodes: Meetup, Space
Blue Link: atPlace
Yellow Action: CreateMeetup
Agent node
Green Deployed target: sideseat

The right panel edits the selected Object’s type, description, and properties (id, title, time, capacity, etc.).

Part 3

Layout · UI tour

Layout

Area	Role
Studio	`platform-web` — node editor, ontology·workflow visualization, save/run/demo/export
Worker / API	Execution, tools, policy; connect via `WORKER_BASE_URL`, etc.
Ontology	Object Type / Link Type / Action — schema, relations, governed mutations
Docker	Role-based runtime/deploy split (`ops` / `backend` / `support`)

Studio UI tour

Studio — this editor
Agents — configured LLM agents (UI may vary by version)
Overview — dashboard
Docs — docs entry

Palette

Section	Nodes	Role
ONTOLOGY	Object Type, Link Type, Action	Schema, relations, governed changes
WORKFLOW	Start, Dataset, Stream / Transform, Agent	Pipeline + LLM blocks
TARGETS	Deployed app, Notify	Deploy targets, webhooks

Canvas & properties

Connect nodes with edges. The properties panel edits metadata for the selection (PROPERTIES — OBJECT, etc.).

Part 4

Local run · architecture · wrap-up

Local run

git clone https://github.com/calicorone/agent-control-plane.git
cd agent-control-plane
cp .env.example .env   # WORKER_BASE_URL, etc.
npm install

Run the platform UI per README (e.g. npm run platform → http://localhost:3010). For Vite HMR, use PLATFORM_API_ONLY=1 npm run platform plus cd platform-web && npm run dev in a second terminal. Ports and scripts: README and docs/LOCAL_SETUP.md.

Architecture in brief

contract vs runtime

Studio          → edit graph·schema·workflow contracts
Worker / models → run Think → Act → Observe
Shared language → Object · Link · Action (see ontology essay)

Stack (short): TypeScript, React, Vite (front end); Node.js, Docker (runtime/deploy); Ollama, Anthropic, etc. per repo config.

Dev notes

Run npm install after clone.
JSON / workflow XML exports help move graphs across environments.
Bearer token: remote Worker or protected APIs only.
Screenshot: static/images/agent-homepage.png (legacy filename).

References

AI Agent Scaffolding & Ontology (KO)
Similar walkthrough: hibiki GPT-2 124M web guide (KO)

In one line

Wire a graph end-to-end, then Execute or export and align with Worker logs—that maps the loop to code and UI fastest.