05 · BUILDERFrom the world of technology and automation
Connect data and judgment into a trading process
You understand systems, integrations and automated processes. Maybe you've already built automations, worked with AI tools, or connected services through APIs and Webhooks.
You may also trade, or your trading experience may still be basic. You want to understand how to build a process where information from the market and the account turns into analysis, a decision and an action.
Reactor is a trading operating system that connects these parts. You can define Agents with roles, data tools and permissions, give them room for judgment, and connect them into a scheduled or event-driven process.
You can start from ready-made prompt examples. The instructions define the task and the way of thinking; the system settings decide the connections, the scheduling and the authority to act.
The examples below show a few ways to use the system.
01
EXAMPLE 1
"I know how to work with AI. How does it get information here that it can act on?"
Reactor connects the AI to tools that bring in data: prices, candles, technical indicators, positions and account details. In the relevant markets, data such as the order book, funding and open interest is also available.
You can define which data a task needs, and check its source and timestamp. The difference between a live price, a snapshot and a closed candle matters especially when the analysis leads to action.
Prices
Candles
Technical indicators
Positions
Account details
Order book · funding · open interest
WHAT YOU'LL SEE
An analysis request, the tool calls and their results, and then the conclusion the Agent built from the information.
Questions that may come up
Is all the data available in every market?
Availability depends on the broker, the asset and the supported tools. We'll show the data available for the task we choose.
How do you know whether a data point fits the decision?
Check the source, the timestamp and the context. You can set requirements for freshness and completeness.
What do you do when data is missing?
Define how the process should respond: report, wait, or not continue. Check the behavior in sample runs.
✓ HOW WILL I KNOW I GOT IT?
You can identify which data the task needs, and what should happen when one of them is missing or stale.
02
EXAMPLE 2
"I'm used to condition-and-action automation. How do you define a process that makes a complex decision?"
An Agent can get a goal, gather information and weigh several factors before deciding. The instructions can set priorities, ruling-out conditions, and a way to handle conflicting information.
For example, an Agent can assess an opportunity and choose between acting, waiting and staying out. Binding limits, such as the task scope and the authority to execute, define the frame it works in.
Act
Wait
Stay out
WHAT YOU'LL SEE
The same task in several market situations, with different decisions and the reasoning for each.
Questions that may come up
What should go into the prompt?
The role, the goal, the information needed, the considerations, the limits of action and the desired output. You can start from a ready-made example.
Will the same prompt always give the same decision?
You need to check consistency in practice. The data, the model and the instructions all affect the result.
How do you choose a model?
You can compare models on representative tasks and look at decision quality, run time and cost.
✓ HOW WILL I KNOW I GOT IT?
You can separate the judgment you leave to the Agent from the limits that must always hold.
03
EXAMPLE 3
"I want to build a full process without loading everything onto one Agent."
You can split the work into roles: scanning, gathering information, the entry decision, position management and reporting.
One Agent's output can feed the next. At each hand-off you define what information is required, what the result means and when to continue. That way you can test each role separately, and the connection between them too.
Scanning
→
Gathering information
→
Entry decision
→
Position management
→
Reporting
WHAT YOU'LL SEE
A short process of detection → analysis and decision → reporting or action, including a case where the process stops.
Questions that may come up
Does the output need to be structured?
Structured output can make passing information easier. Define clear fields and check how the next step uses them.
How do you stop the next step from acting on unsuitable information?
Define input requirements, and the behavior when information is missing or a result doesn't justify continuing.
Can each Agent get different authority?
Yes. You can separate an Agent that only analyzes from one that's authorized to take actions.
✓ HOW WILL I KNOW I GOT IT?
You can describe what each step receives, what it returns, and under which conditions the process continues.
04
EXAMPLE 4
"I already have alerts or a system that produces signals. I want to add analysis and action to it."
With Webhooks you can send an external signal into Reactor and trigger a process from it. The Agent can gather more information, assess the signal and decide how to proceed.
You can also send information out to an execution system, alerts or a dashboard, according to the connection and mapping you defined.
External signal (Webhook)
→
Analysis in Reactor
→
Action or report
WHAT YOU'LL SEE
An incoming signal, the data that was sent, the analysis that was added, and the action or report that followed.
Questions that may come up
Does an incoming signal have to lead to a buy?
You can use it as a starting point for analysis. The Agent can decide to wait or stay out.
How do you map the fields between the two systems?
Define a mapping of the information and check that the asset, the direction and any extra data are interpreted correctly.
How do you check the connection works?
Check the event history, the delivery and the run created by the signal, including handling of errors and duplicate events.
✓ HOW WILL I KNOW I GOT IT?
You can follow one signal from the original system all the way to the result in Reactor.
05
EXAMPLE 5
"The process works technically. How do I check that its decisions are useful?"
Reactor lets you separate the automation's success from the quality of the trading. You can collect signals, replay price movement after them, and examine the decisions and the trade management.
In a real example, a Mapper scanned for Bullish Engulfing every four hours. The candle tool was used to research the movement after the signals. Based on the findings, a 3% stop-loss and a 4.5% profit target were chosen, and a Buyer that buys the signals was run.
As of preparing this example, by the measurement provided, 800 trades closed with a 65% win rate. The Buyer currently doesn't filter with judgment, in order to collect results before testing the next improvement.
4h
Mapper: Bullish Engulfing every four hours
3%
Stop-loss target
4.5%
Profit target
800
Closed trades
65%
Win rate
As of preparing this example, by the measurement provided.
WHAT YOU'LL SEE
The link between the signals file, the buy actions and the results, and how the data lets you test a change.
Questions that may come up
What do you measure beyond the run finishing successfully?
Decision quality, price behavior, profits and losses, costs, and how trades were closed.
Can you also check cases where the Agent didn't act?
You can research collected signals and look at the movement after them, depending on the data available.
How do you add a decision layer without losing the baseline?
Record the version and the decisions, and compare the change against the baseline, accounting for the period and market conditions.
✓ HOW WILL I KNOW I GOT IT?
You can define, separately, the technical success of the process and the success of the decisions it makes.
06
EXAMPLE 6
"The Agent decided to act. How does the decision become an order in the account?"
A trading decision has to become an execution plan: asset, direction, quantity, order type, price and protections.
Reactor lets you build orders, including staged entries and different allocations, according to the supported tools. In the relevant path you can review a preview before execution.
If you're new to trading, we'll also show the difference between the allocated amount, market exposure and the possible loss. Using leverage changes both the exposure and the risk.
Decision
→
Execution plan
→
Orders
→
Actual account state
WHAT YOU'LL SEE
A decision, a trade plan, the orders created, and the actual state of the account after execution.
Questions that may come up
Does making a decision mean the order was executed?
You need to check separately that the order was sent, its status, and the actual execution.
Do all brokers support the same orders?
Capabilities depend on the account, the asset and the broker. We'll look at the relevant execution path.
Where does it make sense to keep human approval?
You can start with approval before action, and consider widening authority after checking the behavior and your requirements.
✓ HOW WILL I KNOW I GOT IT?
You can explain which money action comes out of the decision, and how you verify it was executed as planned.
07
EXAMPLE 7
"I want to know when the process runs, what happened, and how much it costs."
You can run Agents manually, at set times, after another Agent, or in response to an external signal. The schedule should match the pace of the data and the relevant trading hours.
The run record lets you check which model worked, which tools were called, what each tool returned, what failed, how long it took, and what the usage cost was.
Manually
At set times
After another Agent
On an external signal
WHAT YOU'LL SEE
The process schedule, a healthy run and a run with a problem, alongside the tool results and costs.
Questions that may come up
Is a higher frequency always better?
Match the frequency to the task. Re-checking information that hasn't changed can add cost without improving the decision.
How do you find the source of a failure?
Check the run steps and the tool results, then the information passed to the next step.
How do you estimate the running cost?
Measure the cost of representative runs, and factor in frequency, model and the size of the process.
✓ HOW WILL I KNOW I GOT IT?
You can choose a suitable frequency, spot a problem in a run, and estimate the cost of the process.
08
EXAMPLE 8
"I already work in another AI environment and want to use Reactor's capabilities from inside it."
Reactor also supports a connection through MCP, which lets other AI tools access Reactor's capabilities with authorization and subject to controls.
The connection can enable work with market and account data and other tools, depending on the capabilities exposed and the permissions granted.
Webhook
An event that triggers a process
MCP
An AI tool that uses Reactor's tools while it works
WHAT YOU'LL SEE
A task from an external AI environment, the tools available to it, and the information or action obtained through Reactor.
Questions that may come up
What's the difference between MCP and a Webhook?
A Webhook delivers an event that triggers a process. MCP lets an AI tool request the use of tools while it works.
Does the connection automatically grant trading authority?
Access depends on permissions and controls. Check which capabilities were allowed for the specific connection.
How do you choose which connection to use?
By the need: a signal coming from an existing system, passing a result out, or an AI environment that needs to use Reactor's tools.
✓ HOW WILL I KNOW I GOT IT?
You can choose the right type of connection and describe what information and authority it needs.
SEE IT LIVE
Want to go through these examples with us?
In a live demo we'll open the system and walk through the examples that fit you, at your pace.
ANOTHER TYPE?
Not quite you?
Go back and pick the description closest to where you are today.
