Agents Overview
Agents are pre-configured AI assistants with specific instructions, model preferences, and tool sets. Instead of explaining your context every time you chat, create an agent that already knows your workflow.
Why Use Agents
Without agents, every conversation starts from zero. You re-explain your role, your tools, and your preferences. Agents eliminate that overhead.
Each agent carries:
- System instructions -- persistent context the AI follows in every message
- Default model -- the preferred model for this agent's tasks
- Tool access -- which integrations the agent can access
- Color, icon, and avatar -- visual identification in the UI
Example Agents
| Agent | Tools | Purpose |
|---|---|---|
| Code Review Agent | GitHub | Reviews PRs, checks code quality |
| Marketing Agent | Facebook Ads, Google Marketing Suite | Creates campaigns, analyzes performance |
| Project Manager | ClickUp | Manages tasks, tracks progress |
| Content Writer | Google Docs, Drive | Drafts documents, organizes files |
Agent Components
Every agent is defined by:
- Name -- display name shown in the UI
- Icon and avatar -- visual identifiers (a Lucide icon name plus an optional voxel avatar)
- Color -- accent color for the agent card
- System instructions -- the prompt that shapes the agent's behavior
- Default model -- which AI model this agent prefers (or Auto to match the chat)
- Tool access -- which integrations this agent can use
How agents work
In Developer mode, choose the agent from the Agents dropdown in the composer:
- Open the Agents dropdown in the composer area.
- Pick what the chat runs as. The Main agent is single-select; None (generic assistant) clears it.
- Start chatting. That agent's instructions, knowledge, and tools apply to every message.
- Need a different specialist? Pick another one. It takes over from that point and sees everything said so far.
In Simple mode there is no picker: name an agent in your message — @marketing, or a phrase like "use the marketing agent" — and it sticks for that chat. A chip above the composer shows which agent is active, with an × to stop using it.
You talk to one agent at a time -- the one that's active. To combine strengths, work through part of a task with one agent, then switch to another and keep going; the new agent inherits the conversation, so handing off (for example, research with one agent and visuals with another) is just a click.
There's no hidden background delegation. You're always talking to the agent you picked, and you can see exactly what it's doing -- which keeps things predictable and easy to follow.