Getting Started with Agent

Pro+Agent Available on the Pro+Agent tier.

Your workspace knows a lot — every page you’ve written, every connection you’ve linked, every tag you’ve applied. The agent can read all of it, reason about it, and help you develop your world in ways a search box alone cannot. This guide walks you through starting the agent for the first time.

Before you start

The agent requires a configured LLM provider before it can respond meaningfully. If you haven’t set one up yet, see Configure LLM Providers first. You can still start the agent without a provider — it will run in a limited state and surface a “not configured” message when you send your first message.

You also need a workspace open. The agent is scoped to the active workspace and cannot start without one.

Start the agent

Open the agent conversation panel using the toolbar button or keyboard shortcut. If the agent is not yet running, the panel body shows “The agent is not running.” with a Start Agent button.

Click Start Agent. The status dot in the panel header transitions to green and the status label reads “Ready.” The message input and send button appear.

If the panel status reads “Ready” without you clicking anything, the agent was already started from a previous session in this workspace.

Single session per workspace

Inklings runs exactly one agent session per workspace at a time. Starting a second session while one is already running returns an error — this is intentional, not a bug. Stop the current session first if you need to restart with different settings.

Send your first message

Click the message input at the bottom of the panel and type your message. Press Enter to send (or Shift+Enter to add a line break without sending).

The send button becomes a stop button while the agent is generating. The status label reads “Thinking…” and the response appears progressively as the agent streams its reply.

When the response completes, the status label returns to “Ready” and the input re-enables for your next message.

Interrupt a response

If a response is taking longer than you want, click the Stop button (the square icon that replaces the send arrow while streaming). The agent stops at its next checkpoint and a “Session interrupted.” notice appears in the thread. The agent remains running and ready for your next message.

Stop the agent

To stop the agent entirely, use the stop control in the agent panel header. Stopping the agent ends the session. The panel returns to the “The agent is not running.” state. You can restart it at any time by clicking Start Agent again — the harness is fully reusable.

Stopping and restarting does not clear your conversation history from the current session.

Understand agent status

Status	Meaning
Stopped	Agent is not running. No messages can be sent.
Ready	Agent is running and waiting for a message.
Thinking…	Agent is generating a response.
Using <tool>…	Agent is executing a workspace tool.
Error	Agent encountered a fatal error. Restart to recover.

See Also

• Configure LLM Providers — Connect your API key or local model
• Agent Conversation — How the conversation panel works
• Agent Permissions — Control what the agent can read and write
• Agent Tools — What the agent can do with your workspace
• Agent Memory — How the agent remembers context across sessions
• Agent Skills — How skills shape the agent’s behavior