Introduction to Workflow Agents

Workflow Agents lets you build AI agents that do real work with your tools, share them with the people who need them, and run them from the console, a chat, Slack or a schedule. Every agent uses your organization's models and MCP connections, so each model call and tool call goes through the same gateways, guardrails and logs as the rest of your AI traffic.

Go toQuilrAI ConsoleSettingsAI GatewayWorkflow AgentsCreate agent

Start a run

Chat or Start a task

Slack mention or DM

Schedule

Workflow agent

Model from your LLM providers

Tools from your MCP connections

Engine runs the steps

Governed

✓ Tool approvals

✓ MCP Gateway rules and guardrails

✓ Run limits

Result

Markdown, JSON or a table

Posted back where the run started

QuilrAI

Key concepts

Term	What it is
Agent	A published definition: a name and outcome, a System prompt, a model, the tools it may call, an engine and run limits. Each card on My agents shows its Model, Tools and Engine.
Draft	An agent you are still building. Drafts are private to you until you publish.
Revision	Every publish creates a new version. Runs already in progress keep the version they started with, and you can restore an earlier one.
Model	The lead model that writes every answer. Use one of Your models (providers your organization connected to the LLM Gateway) or a Quilr model.
Tools	Tools from your MCP connections, picked one by one. Read-only tools are selected for you; tools that change data are your call.
Engine	The runtime that carries out the steps: LangGraph (recommended), OpenAI Agents SDK, Claude Agent SDK, Google ADK or CrewAI.
Team	Two to six published agents under a lead model that combines their answers into one result.
Run	One chat response or task. Runs show up on the Usage tab.

The page

Workflow Agents has Build a team and + Create agent in the header, and four tabs:

Tab	What it is for
My agents	Your published agents, your drafts and the agents shared with you. Search, switch between Cards and List, and filter with All, Published, Drafts, Shared and Disabled. Connect Slack adds Slack as a tool.
Library	Editable templates to start from. See Library templates.
Chat	Talk to an agent. See Chat and tasks.
Usage	Runs, completion rate, tool attempts and duration per agent. See Usage.

Each agent card has Explore agent, Manage (for agents you own or manage), Start a task and Chat. Drafts show Continue building.

How runs are governed

• Models come from the providers your organization connected to the LLM Gateway, or from QuilrAI-provided models that draw on your shared credit.
• Tools are your MCP Gateway connections. Only the tools you tick are available to the agent, and each person's own connections are used unless you choose otherwise.
• Approvals: by default every tool action waits for the person running the agent. Owners can let shared runs skip that wait, but confirmations your MCP Gateway requires still pause the run.
• Limits: Maximum turns and Run timeout (seconds) stop a run that goes on too long.
• Access is checked again on every run: model access, tool access and who the agent is shared with.

Next steps

• Quick start: build, publish and run your first agent
• Create an agent: every step of the builder
• Chat from Slack: let people talk to an agent in Slack