Agents

An Agent is an AI assistant configured for a specific job. You define what it knows, what it can do, and which AI model powers it. Once active, it can carry out tasks on your behalf using the Skills and Tools you have assigned, and you can chat with it directly.

Agent fields

Field	What it does
Name	A short label that identifies this Agent (e.g. “Project Triage Assistant”)
Description	A note for your team — what this Agent is for
Model	The AI model the Agent uses to think and respond
Max Steps	How many reasoning steps the Agent can take before stopping — prevents runaway loops
Status	Whether the Agent is draft, active, or archived
Usage	Where the Agent is allowed to run — see below
Initial Instructions	The standing instructions the Agent always follows, no matter what task it receives

Agent status

Status	Meaning
Draft	The Agent is being configured. It cannot process live requests.
Active	The Agent is live and available to do work.
Archived	The Agent is retired. It is kept for reference but cannot be activated without being un-archived.

Usage: where an Agent can run

The Usage field controls how the Agent can be started:

Usage	Meaning
Main	You (or a teammate) can pick this Agent and chat with it directly.
Subagent	The Agent isn’t picked directly — it’s only offered as a helper that Polaris (or another agent) can spawn mid-conversation to handle a sub-task.
Both	Either — pickable directly, and available as a subagent.

The Model field

The Model field selects which AI model powers the Agent. GreenVee currently offers a curated set of about seven models from different providers — including several Claude models (Opus, Sonnet, Haiku), GLM, and MiniMax — each with different strengths, speed, and cost. The choice of model affects how the Agent reasons and how it responds.

Note

GreenVee does not require you to understand how each model works. If your team is unsure which model to choose, ask your GreenVee administrator — they will know which models have been configured for your account.

Talking to an Agent

Any active Agent with Usage main or both can be chatted with directly, the same way you’d chat with Polaris. Open Polaris and pick the Agent from the new-chat agent picker instead of using the shared default — you get a normal multi-turn conversation, with its own title and unread indicator, scoped to that Agent’s instructions, Skills, and MCP Servers.

Initial instructions

Initial instructions are always-on guidance written for the Agent. Every task the Agent receives is processed in the context of these instructions.

Use initial instructions to:

• Define the Agent’s role and tone (“You are a project coordinator assistant for GreenVee. Be concise and professional.”)
• Set boundaries (“Only answer questions about open projects. Do not discuss financial data.”)
• Provide background context the Agent always needs

Combining Skills and MCP Servers

An Agent on its own can only reason and respond in text. To make it useful, you give it:

• Agent Skills — named capabilities with their own instructions (e.g. “Triage incoming requests”). The Agent draws on these when they are relevant.
• MCP Servers — external toolboxes that give the Agent real-world actions (e.g. “look up a contact”, “create a record”). An Agent can be connected to multiple MCP Servers at once.

A single Agent Skill can be shared across many Agents. A single MCP Server can be shared across many Agents. You build reusable pieces and assemble them per Agent.

Tip

Start an Agent in draft status. Give it a few Skills and point it at an MCP Server. Test it before switching status to active.

Related pages

• AI Glossary — plain-English definitions for Agent, Skill, MCP, and Model
• Agent Skills — what Skills are and how to create them
• MCP Servers & Tools — how to register a toolbox for your Agent
• Polaris — the shared assistant, and how to start a chat bound to a specific Agent