Agents and professions

An agent is a few choices you make once. Everything else, it does.

Creating an agent

The first agent is created by the setup assistant; the next ones from Nouvel agent in the left column. Each one is defined by:

Name	How it appears in the column, and how sub-agents refer to it.
Model access	By subscription (one of the six vendors) or by OpenRouter key. The exact model is chosen per conversation.
Profession	The prompt and skills it starts with. Four to choose from, below.
Context	Optional: a few lines, some files. Read before every answer.
Working folder	Where shells start and relative paths resolve. Pick it in the explorer, or keep the default.
Scope	Working folder only: file tools refuse any path outside it, and a shell command naming one is refused before it runs. Whole disk: any absolute path.
Tools	Ask before acting: shells, file changes, backtests and sub-agents wait for your consent. Direct execution: they run. Exchange orders always ask, whatever the mode.
Exchange accounts	Which of your accounts this agent may read and pilot. None by default.

The four professions

A profession is a system prompt plus a set of skills, all plain files. The final prompt an agent gets is your profile, its profession, its own context, and a description of its environment: tools, scope, BytNode state, skills.

Profession	Good at	Own skills
Standard	Finance and crypto in general, no speciality. The right first contact.	Reading the market
Analyst	Reads market data and the macro context, derives trends and scenarios, with a confidence level.	Trend analysis, macro context, analysis report
Strategist	Designs complete strategies: objective, entry and exit rules, risk, test plan. Crypto first, open to other markets.	Strategy design, risk management, backtest plan
Quant	Mathematics, statistics and Python: models, backtests and implements algorithms with rigour.	Quant Python, statistical validation, Python backtest

All four share the common skills: how to query BytNode, the shape of a financial answer, trade execution, trade review, and deep research.

Conversations

Open an agent and its conversations are listed. Each conversation picks its model and its reasoning effort in the composer. The + button attaches files (dropped into entrees/ in the working folder, read by the agent’s tools, never pasted into the prompt), takes a screenshot through the browser, or arms deep research for the next message.

At the bottom right of the thread, a gauge shows the context used against the model’s window, tokens per second, and the cost of the turn.