Build a Bitcoin trading agent in Tread
One rule in plain English, handed to Tread's Desk Trader agent: a trigger, a size, a take-profit and a stop. The video shows the whole thing; the five steps below are the same path with the checks written out.
- Platform
- Tread
- Time
- About 15 minutes
- Level
- Beginner
This guide builds a Bitcoin trading agent in Tread in about 15 minutes: whitelist one account, pick the Desk Trader agent, write one sentence with a trigger, a size, a take-profit and a stop, check the four numbers in the draft order, then start it. The video shows the full run.
New guides go to subscribers first. Get the next one by email →
What you need
- A Tread account with at least one exchange account connected and whitelisted for agents (Agents is in beta; you'll accept its terms the first time you open it)
- A rule you can say in one sentence: when, what, how much, where you get out
- A size small enough that a wrong fill is a lesson, not a problem
Steps
Open Agents
Sign in to Tread and open Agents from the top menu. This is the chat workspace, separate from the order ticket. Before anything else, check Settings and make sure the account you want to trade is in the Whitelisted column and nothing else is. Agents can only touch whitelisted accounts.
Pick Desk Trader
The agent list is grouped by job: trading, research, analytics. Choose Desk Trader. It's the one that takes an order brief and executes it; Market Maker and the research agents are for something else.
Meet Desk Trader
Read the agent's opening message. It tells you what it does: asks discovery questions, picks the algo (TWAP, VWAP, iceberg and so on), gates the start on a trigger if you give one, chains a take-profit or stop-loss exit, then monitors the order once live. It also offers sample briefs. Read those; they show the level of detail it wants.
Write your rule
Type the rule as one sentence with four numbers in it. Tread reads the trigger, the size, the take-profit and the stop out of the sentence and shows them back to you as separate fields. If any field comes back empty or wrong, your sentence was unclear; fix the sentence, not the field. Prompt #005 is worth pasting first if you want hard limits in place before the brief.
If BTC drops below $80,000, buy 0.1 BTC with a 4% take-profit and a 2% stop-loss.
Review and go live
The agent builds a draft order and waits. Check every number against what you meant: trigger, size, take-profit, stop, and which account. Only then start it. The agent watches the trigger and runs the order when it hits; you can pause, cancel or change the brief in chat at any time. If it pauses itself in Review mode, it has hit something it won't decide alone. Reply before it does anything else.
Before you rely on it
- Only one account is whitelisted, and it's the one you meant
- The draft order shows the same four numbers as your sentence
- The size is small enough to lose on without caring
- You know where Pause and Cancel are before you press Start
- You've looked at the mission log after the first fill
Used in this guide
Compared in
Frequently asked questions
- What is Tread's Desk Trader agent?
- An execution agent that takes an order brief in plain English, picks the algo (TWAP, VWAP, iceberg), waits for a trigger if you give one, chains a take-profit or stop-loss exit and monitors the order once live.
- Do I need to code to build a trading agent in Tread?
- No. You write one sentence with four numbers in it and Tread reads the trigger, size, take-profit and stop out of it, showing each back as a separate field to check.
- How do I keep a trading agent from touching the wrong account?
- Before anything else, open Settings and make sure only the account you mean is in the Whitelisted column. Agents can only trade whitelisted accounts.
- How much should the first agent trade?
- Small enough to lose on without caring. Know where Pause and Cancel are before you press Start, and read the mission log after the first fill.
Get the next guide first
Hands Off is a free newsletter on AI tools for traders and finance professionals, twice a week. New guides, prompts and shortlists go to subscribers before they go anywhere else.
This guide is an editorial walkthrough, not a tested configuration and not financial advice. Review anything a model produces before you act on it.