Skip to main content
The editor · ships now

Agent Joule.
AI coding on a fraction of the energy.

The reference code editor for energy-optimized compute. A fork of Zed that meters every edit in real joules, then resolves most changes deterministically on your machine, for picojoules. The energy of writing code falls, edit over edit.

~10× lower energy, same result
$0.00 a cached edit
local no phone home
agent-joule · refactor auth module 0.381 J total
  • rename isAuthed → hasSession L1 · template 0.000 J
  • propagate rename to 14 call sites L0 · cache 0.000 J
  • add null-guard in token parser L1 · template 0.001 J
  • extract sessionGuard() helper L0 · cache 0.000 J
  • organize imports, format file L1 · template 0.000 J
  • write session-expiry test (novel) L3 · local model 0.380 J
Five of six closed without a model. The model ran once, for the whole change.

A session, metered. Most edits never reach a model.

The thesis

AI coding can usher in energy-optimized compute

Today, each AI edit is a datacenter round-trip, and energy per change is rising. It does not have to. Three moves turn the curve around, and they build on each other.

01

Energy observability

You cannot optimize what you cannot see. Agent Joule meters every edit in real joules, read from the host’s own power counter. The status line shows exactly what a change cost.

02

Energy-optimized compute

Seeing the cost lets you drive it down. A deterministic-first cascade resolves most edits with lookup, search, and templates, for picojoules. A model is the last resort, not the default.

03

Exponential decrease

The cascade learns. Closed work is cached and retrieved, so a growing share of edits resolve at the cheapest tier, and the energy of writing code falls, edit over edit.

The discipline

AI coding, on a SWaP-2C budget

For decades, anything fielded at the edge was built to a SWaP-2C budget: Size, Weight, Power, Cost, and Cooling, optimized together. Agent Joule brings that budget to the act of writing code.

Size

The compute footprint is your editor, not a cluster.

Weight

No model on the hot path. Most edits never leave the machine.

Power

Every edit metered in joules. The cheap tiers resolve for picojoules.

Cost

About 10× lower for the same result. A cached edit is $0.00.

Cooling

Work that never leaves your laptop never heats a datacenter.

Write code on a fraction of the energy

Open source, a fork of Zed. Source on GitHub →