The terminal and the assistant
Terminal has two places where you talk to a language model. They look similar and they are not the same thing, so it is worth being clear about which is which.
The terminal — the main screen
A prompt. You type what you want and it does it: writes a strategy, compiles a contract, derives a wallet, places a trade, reads your positions. It is the application's main function, not a help window bolted to the side.
It acts. Every action it takes, it prints first:
- A plan before the first step. One line per step, so you see the shape before anything runs.
- Each tool call before its result, so nothing happens between two lines you cannot see.
- A refusal that names the rule it is refusing under, rather than a shrug.
--dryon every command. It performs every read and every simulation, prints exactly what it would write, and writes nothing.
Anything that moves money or deploys code stops and waits for you in a modal. There is no setting that turns that off.
The assistant — the right-hand pane
A panel you open beside whatever you are doing. It reads and explains; it does not act.
Ask it what your licence entitles you to, what a bot did last night, why a trade was refused, what a screen means, what a fee is. It answers from your own data and from the platform's documentation, and when the answer would need an action it hands you to the terminal instead of doing it.
The distinction is load-bearing rather than stylistic. A pane that both explains and acts is a pane where "what would happen if" and "do it" are one keystroke apart.
| The terminal | The assistant | |
|---|---|---|
| What it does | Acts | Reads and explains |
| Can move money | Yes, with a confirmation | No |
| Can write files or deploy | Yes, with a diff and a confirmation | No |
| Where it is | The main screen | The right-hand pane |
| Its model | Assigned in Settings → Models | Assigned separately |
Each surface has its own assigned model, because they are asked for different things. See LLM models.
Neither of them is the communicator. The assistant is a conversation with a model about your own data; it posts nowhere and nobody else reads it. Talking to other people is Channels.
What a local model is, and is not
An 8-to-30-billion-parameter model running on your laptop will structure a strategy, explain a market, read a table and write working code against Terminal's runtime. It will also be confidently wrong about a number or a date.
Terminal is built around that fact rather than in spite of it:
- Numbers come from tools, not from the model. A balance, a price, a position is fetched and shown as fetched. The model does not get to assert one.
- Money moves only through the policy you set — allowed tokens, allowed venues, per-trade and per-day caps — and only after you confirm.
- A generated strategy runs in paper first. Promoting it to live money is a separate, deliberate act.
Type /help and the product tells you this too.