axe
Get Started

Best practices

How operators get the most leverage out of Axe sessions — framing, memory, watches, and closes.

Axe is a session-shaped tool. The operators who get the most leverage out of it treat each session like a real piece of work — with a thesis, a frame, a working trail, and a clean exit.

Frame the session before you open the markets

Use a slash-command opener. /perps-open and /poly-open are not just prompts; they create a session frame Axe can pin memory and watches to. A good opener has three parts:

  1. The marketBTC perps, ETH/USD basis, 2024 election polymarket question
  2. The angle — what you are trying to figure out, not just the asset
  3. The disqualifying conditions — what would make this session over
/perps-open BTC perps; thesis is funding-led mean reversion; I'm out if 4h funding flips negative for two prints in a row.

A framed session beats five free-form questions every time.

Let memory carry context

Axe writes a structured local ledger of every session, watch, runbook, and decision. Future sessions can recall it.

  • Don't re-paste yesterday's analysis. Reference it.
  • Don't re-run the same watch you already configured. Resume it.
  • Don't restate a thesis you already wrote down. Open from it.

If you find yourself rebuilding context, that is a memory bug, not a workflow rule. Tell Axe what you wish it had remembered.

Use watches for events, not for reading

A watch is a tripwire. It should fire when something specific changes, not stream a live feed at you. Good watches:

  • funding > 0.02% for 2 prints
  • BTC perps OI down >10% in 1h
  • polymarket odds on X cross 50%

Bad watches: any watch you find yourself muting.

Run a closing update

End a session with /perps-update or /poly-update. The closer captures what happened, what changed about the thesis, and what the next session should pick up. This is what makes Axe compound — the close is where memory is written.

Treat token earnings as signal, not goal

The reward rail tells you what kinds of work the protocol is paying for. Use it as a feedback loop for the harness, not a target to game. Operators who run real work earn naturally; operators who farm earn worse and learn nothing.

Push back on Axe

If a command feels wrong, say so. If a watch is noisy, kill it. If memory recalled the wrong thing, correct it. Axe gets better when its operators give it feedback — and the protocol's ownership rail is designed so the operators who shape it best benefit most.

On this page