product / agents

Agents review your numbers on a schedule while you trade.

An agent asks your data the same question on a schedule or an event: 'which rule is closest to ending the account', 'what cost the most yesterday'. Answers arrive in the app and in Telegram.

Join the list How it works

Reviews get dropped in the first hard week.

A review matters most after a losing streak, and that's exactly when people stop doing it. Mood doesn't stop an agent: it asks its question on schedule.

the question asks itself

How it works here

Agents run on schedule, and each answer gets checked the next day.

You set the question once

The schedule is written as a plain phrase: 'every weekday at 9:30', 'Fridays after the close', 'when a trade closes'. Rules parse the phrase, and the agent asks again instead of guessing one it didn't understand.

A chief of staff

You write to one agent; it hands the question to the right one and shows who got it and why. The decision model picks the route.

The engine computes the numbers

An agent receives computed metrics, your dossier and your rules. The model takes no part in the maths, and every call stays in the trace.

Ten ready duty agents

A prop-rule watchman, a day review, a repeat hunter, week prep, a cost audit, a pre-open briefing, a playbook audit, a lost-account post-mortem and an auditor of the agents' own advice.

A schedule in plain words or an event trigger
An agent short of data stops and asks
A post-mortem lesson is proposed as a playbook rule
Skills: how trades are reviewed at your desk

The orchestrator: a task split into threads

A big question like 'why did March fail' gets split by the orchestrator into up to six parallel threads. Each thread works with its own context, and the result merges into one answer with project decisions.

The thread plan shows before the run
Each thread is visible on its own, with its own trace
Project decisions are saved and used from then on
Runs in the background; you can close the page

The orchestrator: the plan, the threads and the merge

Squads instead of one adviser

Several narrow agents look at one problem: a month-review team, a prop-account war room, a post-mortem team. Their disagreements stay visible.

Three ready squads for three kinds of review
Build your own squad in the editor
Run on a schedule or with a button
Projects: the files and links a squad relies on

Answers in Telegram

The chief of staff is available in Telegram through the same bot as notifications. Send a question from your phone, and it goes to the right agent.

One bot for notifications and questions
The answer's route shows on the phone too
Scheduled agent answers arrive there too
You link the bot once

A trace of every answer

Every step, every data request, every piece of context. You can't check an answer whose path you can't see.

The log is append-only
You see which data sections went into the answer
The model and its version are recorded with the answer
The cost of the answer in operations

Where it differs

A typical AI assistant waits for a question and answers as well as you phrased it. An agent works without you: the question is set in advance and gets asked in a hard week just as in a calm one.

Check it yourself

Open any agent answer and view its trace: what data went out, at which step, what came back and what it cost. The trace screen is built from the same log as the answer.

The order

From an empty screen to the first review

Switch on a duty agent

Take a ready agent or describe your own question in words.

Set the schedule

As a phrase: 'every weekday at 9:30' or 'when a trade closes'.

Read the answers

In the app or in Telegram. Every answer has a trace.

What it does not do

What this module does not do

An agent doesn't trade. Execution lives in the copier, and the two are kept apart.
Agents need a model: AQREVI's within your plan, or your own key.
Agents come with Pro and above.

Questions

The things that stand in the way

Does this send my trades anywhere?

The answer's context goes to the model provider: the one we use, or the one whose key you entered. Settings show which.

How is this different from chatting with a model?

Chat works while you ask questions. An agent works when you don't, and those are the weeks a journal exists for.

Can an agent act on my account?

No. An agent reads and writes text. Orders come from you, or from the copier once you've switched it on.

What does it cost?

An agent answer costs ten operations. Pro has 2,000 operations a month, Apex 7,500. With your own key on a paid plan, no operations are spent.

A review that doesn't get dropped

An agent asks on schedule, so a hard week gets reviewed just like a calm one.

Join the list

Where to go nextAI: review and check Prop guard