Agents

An agent is an AI worker with a distinct personality, set of tools, and behavioral guardrails.

You can think of an agent as a new team member. You give it a name, tell it what its job is, decide how much independence it gets, and point it at the tools it needs. Once set up, the agent can carry out workflows, answer questions in chat, and learn from experience.

What It Does

• Follows instructions — The agent's role description becomes its core identity. Everything it does is filtered through that role.
• Uses tools — Through connections, an agent can manage issues in Linear, post to Slack, query databases, and more.
• Respects guardrails — Guidelines and risk tolerance settings control what the agent can do without asking you first.
• Learns over time — Memory lets the agent recall preferences, past decisions, and lessons learned from previous runs.
• Runs workflows — You assign an agent to a workflow, and it executes each step according to the instructions.

Key Properties

Property	Description	Options
Name	Display name for the agent	Free text
Role	What the agent does (becomes the system prompt)	Free text
Risk Tolerance	How autonomously the agent acts	low / medium / high
Writing Style	Tone and format preferences	Free text
Preferred Model	Which AI model the agent uses	7 models across 3 providers — see Supported Models
Plan Approval	Whether the agent shows its plan before executing	auto / require_approval
Status	Whether the agent is available for work	active / inactive
Enabled Skills	Which integration categories the agent can access	data, search, code, browser, memory, custom
Description	Short summary of the agent's purpose	Free text
Output Length	How verbose the agent's responses are	concise / standard / detailed
Citation Policy	When the agent includes source citations	always / when_available / never
Restrictions	Behavioral constraints the agent must follow	List of text rules
Max Cost Per Run	Spending limit per execution in USD	Dollar amount (optional)

Understanding Risk Tolerance

Risk tolerance is the single most important setting on an agent. It controls how much the agent can do on its own before checking in with you.

Level	Behavior	Best for
Low	The agent asks before every action. Nothing happens without your explicit approval.	Sensitive workflows — finance, legal, customer-facing communications
Medium	The agent proceeds with routine actions but asks before anything sensitive or irreversible.	Most day-to-day operations — reports, data pulls, internal summaries
High	The agent acts autonomously and only asks when truly uncertain or when a guideline requires it.	High-volume, low-risk tasks — data formatting, internal notifications, log analysis

WARNING

Start with low or medium risk tolerance until you are confident the agent handles edge cases well. You can always increase autonomy later.

Plan Approval

When plan approval is set to require_approval, the agent drafts a plan of action before executing and waits for you to review it. This is especially useful for multi-step workflows where you want to sanity-check the approach before the agent starts working.

When set to auto, the agent moves straight from planning to execution. This is faster but gives you less visibility into the agent's reasoning upfront.

TIP

Even with auto plan approval, you can still review everything the agent did in the run log after it finishes.

Enabled Skills

Skills determine what categories of tools an agent can use. Disabling a skill category prevents the agent from accessing any connection in that category, regardless of what connections exist in your workspace.

• data — Database queries, CRM reads and writes, spreadsheet access
• search — Web search, knowledge base lookups
• code — Code execution, script running
• browser — Web browsing, page scraping
• memory — Reading and writing to workspace memory
• custom — Any custom MCP integrations you have added

What's Next

• Agents vs Workflows — When to chat vs. create a workflow — the role/job mental model
• Workflows — Learn how to give your agent a workflow to follow
• Guidelines — Set the rules your agent must obey
• Overview — See how agents fit into the bigger picture