Memory & automations

The leverage in Axe is not the cockpit. It is what the cockpit captures.

Every session, watch, runbook, and decision is written to a structured local ledger. That ledger is yours: it lives on your filesystem, you can read it, you can export it, you can monetize it on your own terms. And future sessions can recall it.

Memory

Axe maintains layered memory across timescales:

• Session memory — the current thesis, the open watches, the running notes
• Operator memory — your role, your venues, your preferences, your past theses
• Market memory — historical regimes, recurring patterns, the runbooks you trust
• Protocol memory — runbooks and insights shared across the operator network

Memory recall is automatic and explicit. Axe pulls relevant context into a session by default and lets you correct what it pulled.

The session ledger

Every session writes a structured record:

~/.axe/sessions/
  2026-04-27T09-12-04Z_perps-open_BTC.jsonl
  2026-04-27T11-44-30Z_poly-open_election.jsonl
  …

Each line is a structured event — a command, a market read, a watch fire, a thesis update, a close. The ledger is the substrate for memory recall, for provenance, for the reward rail, and for any data product an operator chooses to build on top.

You own the ledger. Nothing leaves your machine unless you ship it.

Automations

Automations are scheduled runs of the work you would otherwise do by hand. The cockpit treats them as first-class objects:

/automations list
/automations add watch_funding_btc every 1h
/automations add poly_resolution_sweep every 30m
/automations pause watch_funding_btc

A good automation is a runbook on a schedule: a session opener, a sequence of reads, a tripwire condition, and a close. Subagents replay them in parallel without holding up your active session.

Subagents

Long-running or parallel work can be delegated to subagents. A subagent gets its own session, its own memory scope, and its own reward rail credit. Common patterns:

• Watch farms — a subagent that holds a dozen tripwires you don't want cluttering your active session
• Investigation runs — a subagent that walks a historical search and returns a written-up summary
• Cross-venue monitors — a subagent per venue that surfaces only events that matter to the active session

Subagent activity flows back into operator memory and into the reward rail. Work done by your subagents counts as your work.

Sharing and shipping

Memory is local by default. Three opt-in shipping paths exist:

1. Runbooks — publish a session pattern to the protocol; other operators adopt it; you earn AXE when it fires usefully
2. Insight markets — package a structured insight (a base rate, a regime call, a watch result) and license it to other operators
3. Personal data products — export your own session ledger to wherever you want; the harness helps you anonymize and structure it

Memory is the substrate. The token rail is what turns it into ownership.