Lesson: Your first conversation and picking a model

You finished lesson 1 with one task in mind. The one you decided was worth handing to AI.

This is the lesson where you actually hand it to AI.

Clawless is open in front of you. The four-step onboarding is done. There is a window, and inside the window there is a chat waiting for you to type. The whole rest of the track sits on top of this one capability: send a message, get a reply, decide which model handled it. Once that loop is real to you, the rest of Clawless is just additions to it.

The lesson takes about eleven minutes to read and ten minutes to do at the same time. You can read first and then try, or read with Clawless open and try as you go. Either is fine.

What you are looking at

When you land in Clawless after onboarding, the window has a few obvious zones.

Down the left side there is a vertical list of avatars. That is the agent rail. Each avatar is a different agent, each one set up for a different kind of work. The one selected by default is called Assistant, which is exactly what it sounds like, the all-purpose one to use when you do not know who else to ask. A later lesson introduces the other agents and when to reach for which.

The middle of the window is the chat. The header at the top shows the agent’s name, the conversation title, and the current model. The message area in the middle is empty for now; it will fill up as you and the agent go back and forth. The input box at the bottom is where you type.

Below the input box there is a row of small icons and one dropdown chip. That row is the dock row. It holds the controls you reach for often: a New conversation button, a status dot that tells you whether you are online, a slim Skills shortcut, a slim Tools shortcut, a Commands shortcut, and most importantly for this lesson, the model picker. The dropdown chip with a model name on it (“Claude Opus 4.6” or whatever you set during onboarding) is the model picker. We come back to it in a minute.

Far left, the narrow strip of icons is the navigation rail. Those icons open the panels for Tools, Skills, Memory, Channels, Cron, Logs, Usage, and Settings. You do not need any of them for this lesson.

That is the whole tour. Four zones: the agent rail on the left, the chat in the middle, the dock row below the input, and the navigation rail at the far left.

Send your first message

Click into the input box at the bottom of the chat. Type the question you brought from lesson 1.

If you did not bring one, type this:

Help me write a short status update to my team about a project that is on track but had two small blockers this week. Keep it three sentences. Confident tone.

Press Enter.

Two things happen at once. Below your message, a short status line appears: “Thinking”, “Reading file”, or something similar. That is the activity indicator; it tells you what the agent is doing in real time. Above it, the agent’s reply starts to stream in word by word. You do not have to wait for the whole reply before you start reading.

If the reply is heading somewhere you do not want, you can stop it. Where the Send button used to be, you will now see a square icon. That is the Stop button. Press it and the agent stops streaming.

If the engine takes a beat to catch the stop request, you will see a small “Stopping…” line. If after about ten seconds it has still not stopped, a red banner with a Force Stop button appears. Use Force Stop only as a last resort. Most of the time, regular Stop is instant.

When the reply finishes, code blocks have a Copy button in the corner, tables render as real tables, links are clickable. You read the reply, and if you have a follow-up you type it into the input box and press Enter again. The agent remembers everything in the current conversation, so you do not have to re-paste context every time.

The model picker

Look at the dock row, just below the input box. Find the dropdown chip with a model name on it.

That is the model picker. Click it.

A short list drops down, grouped by provider: Anthropic models in one group, OpenAI models in another, Google in another, and so on (the exact lineup depends on which provider keys you saved during onboarding). The default selection is whichever provider’s key you set up first.

Two things are worth knowing about the list.

First, it is a curated short list, not the full catalog of every model every provider ships. Most providers have dozens of models, and most of the differences between them do not matter for most of the work you will do. Clawless picks a small handful per provider that cover the useful range, capable for hard work, fast for quick work, cheap for high-volume work. The default short list is enough for almost everything.

Second, if the model you want is not on the short list, you can still use it. The picker also accepts a provider-prefixed name typed straight into the input. The format is the provider slug, a slash, and the model name. The docs give the example below:

openai/gpt-5.1-mini

Type that pattern (provider slug, then slash, then model name) into the picker and Clawless uses that model even if it is not on the visible short list. You only need to do this when you have a specific reason; for the first dozen conversations, just pick something from the short list.

The default rule of thumb when you are starting out: pick the biggest, most capable model from whichever provider you trust most. That is the easiest way to see what AI can do well. Lesson 3 of Track 22 (“Choosing your model and the effort dial”) goes into the cost-and-capability arithmetic in detail; the short version is that smaller models are dramatically cheaper, and after a few conversations you will start to notice which tasks the small models handle fine and which need the big ones.

Switch models in the middle of a conversation

This is the load-bearing feature of the picker and it is worth a section of its own.

You can switch models at any point in a conversation. The new model picks up where the previous one left off; it sees the whole conversation history and continues. There is no rule that says “this conversation is locked to one model.” There is no per-conversation lock at all.

Here is when that matters in practice.

You ask a hard question that benefits from a heavyweight model. You let Opus or GPT-5.1 or Gemini 2.5 Pro draft a careful answer. Then you have ten small follow-ups about the draft, none of them as hard as the first question. You switch to a smaller, cheaper model for the follow-ups. The conversation continues as if nothing happened. The cost per message drops by a factor of five or ten.

The reverse also works. You start with a small fast model for an easy task. Halfway through, the task gets harder than you expected. You switch to a bigger model and the conversation continues. The hard part gets the right tool.

Costs follow the model you are using at the moment you send each message. Switching does not retroactively cost anything; it just changes the meter going forward. That is the whole rule.

What happens when you switch agents

Each agent has its own preferred model. When you click a different agent on the left rail, the model picker updates to that agent’s preferred model. This is not a glitch; it is intentional. The Researcher agent probably wants a different default than the Writer agent.

You can still override it for the current conversation in the picker. If you want to change an agent’s preferred model permanently, you edit the agent itself; a later lesson covers the agent settings.

The other thing worth knowing is that each agent keeps its own conversation history. When you click between agents, you do not lose your place in either; you swap between two parallel conversations. That makes it easy to run a quick task on the Assistant while a longer Researcher session is still in progress in the background.

A few small things you will notice

The model name in the conversation header (the very top of the chat area) shows the current model. If you switch models mid-conversation, the header reflects the model you have selected.

The Usage dashboard, on the navigation rail at the far left, breaks down costs by provider, model, and conversation. After your first few conversations it is worth opening once just to see what each model costs you in practice. Reading is more durable than guessing.

Slash commands are a power-user shortcut you can ignore for now. Typing a forward slash at the start of the input opens a small menu of commands, including one to switch models without leaving the keyboard. You do not need them for the first hour with Clawless. A later lesson picks them up.

Common first-day surprises

Four things people often notice on the first day. Knowing them in advance saves frustration.

The first message of a session feels slow. The engine warms up on the first message, and the AI provider sometimes has a cold-start delay on its end. Every subsequent message in the same session is faster. This is normal.
A small status indicator appears below your message while the agent thinks. That is the activity indicator, not a problem. It tells you what is happening: thinking, reading a file, searching the web, calling a tool. If you ever wonder whether the agent is stuck, the activity indicator is your answer.
Switching to a different agent loads a different conversation, not a continuation of the current one. Each agent has its own thread. That is the right design for handing different kinds of work to different specialists, but it surprises people on the first day. If you want to continue the current topic, do not switch agents; keep going in the same one.
The model picker shows different defaults for different agents. Each agent has its own preferred model. Clicking between agents looks like the picker is randomly changing; it is not. The picker is following the agent.

What you should remember

Your first message is just typing and pressing Enter. The chat surface from top to bottom is header, message area, input, dock row. The dock row is where the model picker lives.
The model picker is a curated short list grouped by provider. Pick whatever looks right; the default is fine for almost everything. If you need an off-list model, type the provider-prefixed name into the picker.
You can switch models mid-conversation, and the new model continues where the previous one left off. There is no per-conversation model lock. The cost meter follows the model you are using at the moment you send each message.
Each agent has its own preferred model and its own conversation history. Switching agents swaps both. The chat does not get lost when you move between agents.

In the practice that goes with this lesson, you will send three real messages, switch models once mid-conversation, and try one off-list model. Twenty minutes start to finish. That is the foundation for everything else in this track.