Core Concepts
Agent Studio is Moveworks’ platform for building AI-powered plugins that automate business processes. You define what your plugin does — the reasoning engine handles how to execute it at runtime.
This page maps the core building blocks and how they connect. Read it first, then dive into the sections that match what you’re building.
How a Plugin Executes
Every plugin follows the same lifecycle: a trigger fires, the reasoning engine selects a plugin, slots are collected, actions execute, and a response is returned.
Architecture Map
The diagram below shows how every core concept in Agent Studio relates. The reasoning engine sits at the center — your plugin configuration tells it what to do, and it orchestrates everything else.
Key relationships
- Triggers determine how a plugin activates. User utterances invoke Conversational Plugins. Webhooks and schedules invoke Ambient Agents.
- The Reasoning Engine is the runtime that powers every plugin. It retrieves and ranks plugins, collects slots, resolves values, plans action execution, and stores results in the Data Bank. Policies are the decision layer — they determine each next step (collect a slot, call an action, respond to the user).
- Slots are the inputs your plugin needs. Slot Resolvers convert natural language values (“White Fox conference room”) into system IDs (
R-3824). - Actions are the operations your plugin performs. Five types exist, each for a different use case.
- The Data Bank stores all data generated during execution — slot values, action outputs, and system metadata — so downstream steps can reference upstream results.
- Connectors authenticate and route requests to external systems (ServiceNow, Jira, Salesforce, your own APIs).
Core Concepts Reference
The hierarchy: Assistants contain Agents, Agents use Plugins. Start here to understand what you’re building.
Plugins triggered by user messages. Covers natural language triggers, slot design, conversation processes, and reasoning instructions.
Plugins triggered by system events (webhooks) or schedules. Run without user interaction.
The runtime that powers your plugins: Plugin Retrieval, Planner, Policies, Slot Resolvers, and the Data Bank.
The centralized data store for plugin execution. Access slot values, action outputs, and system metadata via data.* paths.
The type system for slots and action inputs: strings, numbers, booleans, lists, objects, dates, and structured types.
How to provide source attribution in plugin responses so users can verify the information.
How the reasoning engine interprets and presents tabular data returned by actions.
Recommended Reading Order
If you’re new to Agent Studio, read in this order:
- LLM Fundamentals — how LLMs predict tokens, context windows, and attention (explains why every design pattern exists)
- Assistants, Agents, & Plugins — understand the hierarchy
- Conversational Plugins — the most common plugin type
- Designing Conversation Processes — the best practices guide (required reading)
- Data Bank — how data flows between steps
- Actions overview — what operations are available
- Agentic Automation Engine — how the runtime works under the hood