Skip to main content
An agent is a configurable AI assistant. You define its behavior with instructions, choose which tools it can use (via MCP servers), and optionally attach resources and workflows.

​
What an agent is

  • Instructions — The system prompt. Defines behavior, tone, and expertise.
  • MCP servers — Tools from integrations (Google Drive, Gmail, Slack, Salesforce, and many more). You select which servers the agent can use; it discovers and calls tools as needed. See MCPs for details.
  • Resources (optional) — Datastores, filestores, and knowledge bases. You grant read/write/delete per resource.
  • Workflows (optional) — Existing workflows the agent can invoke for multi-step processes.

​
Where you can use agents

  • Web chat — In the Pinkfish app
  • Slack — Connect an agent to a Slack workspace
  • Microsoft Teams — Connect an agent to a Teams tenant
  • API — Use agent API keys for programmatic access. See Agent Management.

​
Agent types

  • Chat — Standalone conversational agent
  • Workflow — Agent bound to a specific workflow

​
Key capabilities

  • Custom instructions — Define behavior and tone
  • MCP tools — Access tools from 100+ integrations; authenticate once in Integrations, agents use your connections
  • Workflow orchestration — Attach workflows for specialized processes
  • Sharing — Share agents with your org via ACLs (integrations do not share; recipients connect their own)

​
Agent Memory

Agent Memory

Store and recall user preferences and context across conversations and sessions.

​
Getting started

  1. Navigate to Agents → Create New Agent
  2. Instructions — Name, description, and system prompt
  3. MCP servers — Select which servers (and thus tools) the agent can use
  4. Resources (optional) — Add datastores, filestores, or knowledge bases
  5. Workflows (optional) — Attach workflows from your library
  6. Save and share — Deploy with your team