Agentic Workspace Overview
The Agentic Workspace is the section of your project where you build, test, and manage AI agents. It lives in the project sidebar under Agentic Workspace and contains three tabs: Agents, MAS Systems, and Agent Personas.
What you can build
Task Agent
A Task Agent is a graph-based workflow agent. You build it on a visual canvas by connecting nodes — each node performs one action (call an LLM, query a table, branch on a condition, etc.). Task Agents are deterministic: they follow the graph exactly as you wired it.
Use a Task Agent when you need a predictable, auditable pipeline — customer onboarding, data enrichment, approval workflows, RAG-powered Q&A.
AI Agent
An AI Agent is an LLM-driven agent that uses tool calling and autonomous planning. Instead of a fixed graph, you give it a goal, a set of tools, and policies. The LLM decides which tools to call and in what order.
Use an AI Agent when the problem is open-ended — research tasks, multi-step reasoning, complex information retrieval where the path isn’t known in advance.
MAS (Multi-Agent System)
A Multi-Agent System coordinates two or more agents working together on a shared goal. You choose an orchestration pattern and assign agents to roles.
Available patterns:
| Pattern | How it works |
|---|---|
| Router Specialists | A Router agent dispatches tasks to specialized agents |
| Planner-Executor-Verifier | A Planner breaks work into steps, Executors run them, a Verifier checks results |
| Hierarchical | A Manager delegates to sub-managers or workers |
| Debate-Judge | Agents debate, a Judge picks the best answer |
| Map-Reduce | Work is split (mapped) across agents, then combined (reduced) |
| ReAct | An Executor agent reasons and acts in a loop |
MAS requires at least 2 agents and supports up to 20 agents per system.
Agent Personas
A Persona defines the personality, tone, and behavior guidelines for an agent. You can assign a persona to any agent — it shapes how the agent communicates with users. Personas include instructions on voice, vocabulary, response length, and guardrails.
The agent lifecycle
Every agent follows the same lifecycle:
- Create — Pick a type (Task or AI), name it, define its goal.
- Build — For Task Agents: wire nodes on the canvas. For AI Agents: configure tools and policies.
- Test — Run the agent in a chat session. Inspect traces, replay, debug.
- Version — Save a snapshot. Compare versions. Promote to production.
- Deploy — Push to a Preview or Production environment.
- Monitor — Watch runs, costs, errors. Iterate.
Task Agent canvas at a glance
The Task Agent canvas is a full-screen visual graph editor (the Agent Builder). You drag nodes from the palette, connect them with edges, and configure each node via its settings panel.
Node palette categories
| Category | Nodes | Purpose |
|---|---|---|
| Flow Control | Start, End, Parallel, Merge | Entry/exit points, parallel branching |
| Actions | Agent Call, Tools Call, Quick Prompt | Sub-agent invocation, tool execution, LLM calls |
| Logic | Condition, Switch | Conditional branching and multi-way routing |
| Loops | Loop | Iterate over arrays or repeat until a condition |
| Data | Query Table, Insert Row, Update Row, Delete Row | CRUD operations on Data Vault tables |
| RAG | Search Content Store, Reindex Content Store | Vector search and re-indexing |
| Advanced | Code | Custom code execution |
The Start node defines the workflow entry point. Configure its Input Variables to pass data into the workflow from external callers.
What’s next
- Create a Task Agent → — Step-by-step creation wizard
- Canvas tour → — Toolbar, palette, inspector, run panel
- Node reference → — Detailed docs for every node type
- Build your first Task Agent → — End-to-end cookbook