Skip to main content

Agents

Overview

Agents are specialized AI assistants you can select in Chat. Each agent bundles instructions and curated context so the assistant responds with a consistent “lens” (for example: pipeline analysis, onboarding support, or renewal risk).

Usage

General usage

  • Select an agent to switch the assistant into a specialized mode with pre-configured context and behavior.
  • Use “No agent” when you want the default assistant behavior.

Creating an agent

Create agents from the workspace Agents page. You can start from a template or from a short description, then refine the configuration.

What you can configure:

  • Identity

    • Name and optional description (what the agent is for)
    • Category (for organization and discoverability)
    • Icon (so it’s easy to recognize in Chat)
  • Instructions

    • The agent’s instructions define how it should behave (priorities, style, and how it should approach analysis).
  • Context (what the agent “knows” by default)

    • Memories: pre-selected knowledge the agent should always have available.
    • Prompts: reusable starting points the agent can expose for common tasks.
    • Document folders: default folders that the agent can search within when answering.
  • Tools

    • Agents can define which tools are enabled for the conversation (for example, web search or code execution) so the behavior is consistent for the task.

Using an agent in Chat

  • How to use an agent

    • In the chat composer, open the Agent selector.
    • Choose an agent (or choose No agent to return to the default assistant).
    • Once selected, Chat uses the agent’s instructions and configured context.
  • How memories work with agents

    • Default mode (no agent): Chat can automatically recall relevant memories for each question and build an auto-updating conversation context (you can lock it).
    • With an agent selected: the agent’s pre-configured memories become the base context, and Chat does not automatically expand memory context on its own.
  • Tools and context are managed by the agent

    • When an agent is selected, the conversation’s tool settings and default document folders follow the agent configuration.

Examples

  • Renewal Risk Agent

    • Purpose: summarize churn risks, surface drivers, and recommend next actions.
    • Best for: account reviews, renewal prep, and weekly risk reporting.
  • Pipeline Hygiene Agent

    • Purpose: spot inconsistent fields, stale opportunities, and missing next steps.
    • Best for: sales ops checks and forecasting prep.

Features

FeatureWhat it does
Specialized behaviorUses agent-specific instructions so the assistant stays consistent in tone, priorities, and approach.
Curated contextStarts with agent-selected memories and prompts so responses are grounded from the first message.
Scoped sourcesCan include default document folders so the agent can search within a known set of documents.
Tool settingsCan manage which tools are enabled for the conversation (for example, web search or code execution).
DiscoverabilityAgents can be categorized, searched, and surfaced by usage (most-used agents show up first).