Terminology

The terms below are listed alphabetically. Each links to the page that explains it in full.

Agent. A form-based AI assistant: a model, instructions, tools, memory and a test chat. It is listed under Agents in the sidebar. See create an agent.

AI Agent (step). The step that does the thinking in an agent. It calls a language model and the tools you gave it. See AI Agent.

Chat history (setting). The agent setting that controls how many earlier messages in a conversation the agent sees when it answers. Not to be confused with the History page. See memory and chat history.

Chatbot. An agent or workflow published as an embeddable or public chat. See create and embed.

Credits. The balance a workspace spends when its agents and workflows use AI models. See credits.

Deep agent. A switch on the agent that gives it a file system, a planning tool and one helper sub-agent. Use it for long, multi-step tasks. See preferences.

Draft. The version of an agent or workflow you are editing. The test chat runs the draft. See how AI Studio works.

History (tab). The tab of the agent builder panel that lists the draft and earlier versions, with a Restore button. See share and publish.

Integration. A connection from your workspace to another service, such as Slack or Gmail. See connecting accounts.

Knowledge. The documents, crawled pages and questions and answers an agent can search. See knowledge.

MCP. Model Context Protocol, the standard way an AI client connects to tools. FlowHunt can act as an MCP server for your workspace, and an agent can use other MCP servers as tools. See MCP.

Memory. Persistent notes an agent can store and look up across conversations, grouped in categories. See memory and chat history.

Published. The live version of an agent or workflow, created by Publish changes. See share and publish.

Run. One execution of an agent or workflow, started by a trigger. See testing and run logs.

Session. One conversation with an agent. The builder keeps your test chats as sessions. See sessions and testing.

Step. One unit of work in an agent or workflow, such as a trigger, a model call or an output. See the step reference.

System message. The instructions that tell the agent its role and how to answer. Set it in Model & instructions. See preferences.

Template. A ready-made agent or workflow you copy into your workspace. See library and templates.

Tool. A step an agent may call on its own while answering, such as a retriever or an integration action. See agent tools.

Trigger. The step that starts a run, for example a chat message, a schedule or a webhook. See workflow triggers.

Workflow. A canvas of steps you wire together yourself. It is listed under Workflows in the sidebar. See canvas basics.

Workspace. The space that holds your agents, workflows, knowledge, chatbots, team and credits. See team and roles.