Talking to Your Agent

This is the core Taurus workflow: you send messages, the agent works, and you iterate together. Everything happens inside runs — discrete units of work that hold a conversation with your agent.

In AI chat apps, this would be analogous to conversations. In coding tools like Claude Code or Codex, these are often called sessions. We call them runs.

Each agent has its own list of runs.

Starting a run

After onboarding, you'll see the first run where the agent greets you. You can continue chatting in that first run.

And when you need a new conversation — a fresh context — click New Run in the header to start with a clean slate.

As you become a power user, you would learn to switch between different runs and have multiple conversations -- the same agent can do several things at once by having multiple active runs.

Sending messages

Just type your message as you usually do in messaging apps. Enter to send, Shift+Enter to insert a newline.

Later, you'll learn how to attach files, images, and use voice input (the microphone button).

Reading agent responses

Agent messages stream in real time — you'll see text appear as it's generated.

Agents do more than just talk. As the agent works, you'll see tool calls appear as collapsible blocks inside the response: file reads, shell commands, web searches, code edits, and more. Click any block to see the full input and output.

Run controls

While a run is actively generating, a floating toolbar appears at the top with a Stop button. Click it to halt the agent mid-work. You can always continue the run later by sending another message.

There is also a Stop All button on the agent top bar to stop all runs of that agent.

Tip: You can send messages to the agent even mid-turn, if you want them to correct course or explain something before continuing.

Tip: Sometimes you don't want the ugly consequences of clicking Stop mid-run and interrupting a tool call that might be in flight. You can send a message like "please pause", and it will be inserted into the agent's context immediately after the current tool call finishes.

Message actions

Open the message menu when you want to act on a specific point in the conversation. The most useful action is usually Fork: it starts a new run from that message so you can try a different direction without losing the original path.

Compaction

Long runs eventually fill context. When that happens, use the context ring or the /compact command to shrink the conversation and make room for more work.

If the task is still the same, compaction is often the right move. If the task has drifted, start a new run instead.

What's next

Now that you know how runs work, head to Creating & Configuring Agents.