The agent copilot

The copilot is an AI assistant for your team, inside the workspace. It can answer questions about your operation and propose actions, always within the limits of the signed-in member's permissions.

Reading your workspace

The copilot can look across your workspace to answer questions (for example, summarizing open conversations, surfacing feedback trends, or finding an article). Every read is double-gated: it runs under your workspace's data isolation, and it is checked against the same permission policy the member's own clicks go through. The copilot can never read anything the signed-in member could not open themselves; a denied request comes back as a plain refusal, not an error.

Proposing actions (confirm first)

The copilot can also propose changes, but proposing is not doing. For a write action it presents what it intends to do and waits for you to confirm before anything happens. Actions it can propose include:

• Closing or reopening a conversation.
• Assigning a conversation to a teammate.
• Adding an internal note or an existing tag.
• Changing a feedback post's status.
• Publishing an article or creating an article draft.
• Creating a feedback post.

Each proposed write is checked against the same permission the manual action requires (for example, publishing an article needs help-center management), and it is clearly flagged when it has a customer-visible effect. The copilot never reports an action as done until the confirmation actually completes.

Configuring the copilot

An admin can disable specific copilot tool groups for the workspace (for example, turning off help-center writes) from the AI settings. The copilot uses the same provider and models you configured for the rest of Hark's AI.

The docs assistant

Separately from the workspace copilot, these product docs have an "Ask the docs" assistant that answers questions about Hark itself from the documentation you are reading now. It is a platform-level feature and does not touch your workspace data.