gridhour-au

When is electricity in the Australian National Electricity Market (NEM) green and cheap? gridhour-au shows the next 48 hours of estimated carbon intensity and wholesale spot prices for your NEM region on one timeline in the terminal, then picks the best time to start the washing, charge the car or kick off a batch job.

$ gridhour-au --line
⚡ 247g · 5c · green now

It is an Australian port of gridhour, which does the same for Great Britain. The interface, jobs and outputs are the same; the data and the way carbon is worked out are not (see Where the numbers come from).

What you see

• The chart. Estimated carbon intensity, in grams of CO₂ equivalent per kilowatt-hour (gCO₂e/kWh), rises above the time axis and the regional spot price hangs below it. Green is clean or cheap, red is dirty or dear, and blue is a negative price. Bars taller than 50 cents per kilowatt-hour (c/kWh) are cut so that one price spike doesn't flatten the rest. The hours already gone are dimmed.
• Generation mix. Coal, gas, hydro, wind, grid solar, rooftop solar and batteries as shares of your region's supply, on the same time axis.
• Best time to run. One row per job. Each row is shaded by how good every start time is, and the bright block is the best window, with its average carbon and price next to it. The status line tells you how much that saves against starting right now.
• A cursor runs straight down through all of it. Move it with the arrow keys to read any half hour; Enter jumps to the selected job's best window.

Times are shown in your region's local time, daylight saving included: Sydney is on Australian Eastern Daylight Time (AEDT) in summer, while Brisbane stays on Australian Eastern Standard Time (AEST) all year.

Jobs are ranked by carbon, by price, or by both (each normalised over the 48 hours and averaged). w switches between them.

Deadlines and jobs that can run in pieces

Washing      2h
EV charge    4h by 07:00 split
Battery      3h by 16:00 split

• by 07:00 means the window has to finish by the next 07:00, local time. An amber ┤ marks the deadline on the job's row. If there isn't enough forecast before it, gridhour-au says so rather than suggesting a time after it.
• split lets it pick the best half hours before the deadline in any order. It only splits when the pieces really are better than one continuous block.

Type them in the add box (a), or select a job and press b for the deadline and s to allow pieces.

Scrubbed to tomorrow	Compact layout for a tmux split

Where the numbers come from

Everything comes from the public NEMWeb reports of the Australian Energy Market Operator (AEMO). No account or key is needed.

Predispatch runs to 04:00 (NEM time) at the end of the next trading day, so it covers roughly the next 16 to 40 hours depending on the time of day. AEMO's Short Term Projected Assessment of System Adequacy (STPASA) and its seven-day predispatch (PD7Day) fill the rest of the 48 hours.

Price is the regional reference price (RRP) AEMO forecasts, divided by ten to give c/kWh. That is the wholesale spot price, before network charges, the retailer's margin and the goods and services tax (GST). If your retailer passes the wholesale price through to you, its rises and falls are what you pay; on a flat or time-of-use tariff they aren't, and --mode green (or w) ranks by carbon alone.

Carbon is an estimate. AEMO does not forecast emissions intensity, so for each half hour:

1. Grid wind and solar, and battery discharge, come from predispatch, as AEMO expects them to be dispatched. Rooftop solar comes from AEMO's rooftop forecast.
2. The rest of the region's generation (forecast generation, less grid wind, solar and batteries) is taken to come from the region's other units (coal, gas, hydro, diesel, bioenergy), in the proportions and at the emissions intensity they are running at right now. That intensity is each unit's latest 5-minute reading from AEMO's supervisory control and data acquisition (SCADA) system, multiplied by AEMO's emissions factor for that unit.
3. Battery discharge is counted at the intensity of the half hours the batteries charge in (see below).
4. Intensity = emissions of that fleet and the batteries / (fleet + batteries + grid wind + grid solar + rooftop solar).

Past the end of predispatch

For the rest of the 48 hours (as much as 32 hours of it, in the early afternoon), AEMO publishes no market forecast, only its STPASA reliability study. That has the region's demand and the wind and solar output forecast to be available, but not how much the region will generate, so two things are estimated:

• Generation. A region rarely generates exactly its demand: it exports when wind and solar run ahead of it and imports when they don't, and batteries and pumped hydro take some on top. Across the predispatch run, gridhour-au fits a straight line relating that gap (generation less demand) to how far wind and solar exceed demand, and uses it to turn STPASA's demand into generation. Wind and solar beyond what that generation leaves room for are taken as curtailed, and the fleet is never set below its lowest output in predispatch, because coal units keep running and export on a sunny afternoon.
• Wind and solar. These are the available-capacity forecast, the unconstrained intermittent generation forecast (UIGF). STPASA also has "cleared" figures, but they come from the reliability run, which dispatches only what it needs: across a week of New South Wales (NSW) noons they showed 1.6 to 2.0 gigawatts (GW) of solar against 2.8 to 5.0 GW forecast.

To test the line, it was fitted to the first half of a predispatch run and used to predict the fleet's output in the second half. On both test days (9 and 27 October 2026), it cut the error in NSW, Queensland and Victoria compared with taking generation as equal to demand: by 34% to 70%, except in Victoria on 27 October, where the cut was 7%. In South Australia it was better on one day and worse on the other. The fleet there is small (often under 200 megawatts, MW) and swings with the interconnectors, so expect South Australia's later hours to jump about more. Where the two reports meet, the NSW, Queensland and Victoria estimates matched to within 2% on both days. Taking generation as equal to demand had produced steps of up to 5% on those days, and a review on 16 October found one of 11% in NSW.

These hours are still rougher than the predispatch ones. They assume the export pattern of the next day and a half continues, and they know nothing about outages or interconnector limits.

The estimate leaves some things out, so treat the numbers as a guide to which hours are cleaner rather than as exact figures:

• Interconnectors. The estimate is the intensity of what is generated in your region. Imports and exports are not followed, so Tasmania reads close to zero even when Basslink is bringing in Victorian coal power.
• The fleet's mix changes. Gas peakers that start for the evening peak, or coal units coming back from an outage, only show up once they are running.
• Beyond predispatch. Generation is estimated from demand, as described above.
• Batteries. They are shown as their own share of the mix (in Queensland at 19:00 NEM time on 26 October 2026, they supplied 1.3 GW of 8.3 GW). A battery makes nothing: it returns energy the grid made earlier, so its output carries the carbon of when it was charged. That is the estimated intensity of the predispatch half hours in which the region's batteries charge, weighted by how much they take in. It covers the last 24 hours too: as each predispatch half hour passes, gridhour-au keeps the last forecast made for it, because batteries often discharge more than the current run shows them charging, drawing on what they held when the run began. In six runs on 26 and 27 October 2026, the charging intensity from the runs alone came to 2% to 44% of the fleet's intensity (lowest in South Australia, highest in Victoria). Adding the previous day raised it in 23 of the 24 combinations of region and run, and changed no half hour's estimate by more than 28 g. Two things are assumed:
  • Until gridhour-au has been running for a day, and for energy stored for longer than that, charging it cannot see is assumed to resemble the charging it can. Filling in that day from older predispatch runs on a fresh start doesn't help: across 6 runs and 4 regions on 26 and 27 October 2026, using two or four older runs (6 to 24 hours back) left the result as far from the full-day figure as using the current run alone, typically 8 to 16 g in a half hour and at most 27 g. Forecasts made hours ahead expect batteries to charge in the clean middle of the day, as the current one does; only the final forecast for each half hour shows them also charging in dirtier hours, and collecting those would take one download per half hour.
  • Past the end of predispatch there is no battery forecast, so battery discharge there is counted as part of the fleet's share, at the fleet's intensity.
• Small generators. Small non-scheduled generators are not in the forecasts.
• Units without an emissions factor. A few running units are missing from AEMO's list (in October 2026, two Braemar gas units and two Kareeya hydro units in Queensland), so the fleet's intensity is worked out without them. In four snapshots, this made Queensland read 0.4% to 1.3% high.

"Green" means the estimate is under 250 gCO₂e/kWh (graded "very low" under 100, "low" under 250, "moderate" under 450, "high" under 650 and "very high" above that). These are fixed bands, so in Victoria and Queensland green hours are mostly sunny middays, while in South Australia and Tasmania most hours qualify.

On the first run, the two hours before now are empty. After that, each half hour that passes keeps the predispatch and rooftop forecasts made just before it. Its carbon is still worked out with the fleet's current intensity, so it can shift a little from one refresh to the next. If a refresh gets a new STPASA run but not the predispatch run that goes with it, the hours between the two come from the previous STPASA run rather than being left empty.

Reports are reduced to a few numbers per half hour and cached in ~/.cache/gridhour-au. A cached report is checked for a newer one after 30 minutes (2 hours for the 5 MB PD7Day report, a day for the emissions factors), and the file is only downloaded again if NEMWeb lists a newer one. If the network is down, the app keeps showing the cached data and says how old it is. GRIDHOUR_AU_OFFLINE=1 stops all network access.

gridhour-au is not affiliated with AEMO.

Install

You need Python 3.11 or newer, and a terminal with truecolor and a UTF-8 locale. It runs on Linux and macOS; on Windows, use the Windows Subsystem for Linux (WSL). The simplest way to install it is with pipx:

pipx install gridhour-au                                 # from the Python Package Index (PyPI)
pipx install git+https://github.com/vlasky/gridhour-au   # or straight from GitHub
gridhour-au 2000

From a clone, pipx install . does the same. To run it without installing anything, use python3 -m gridhour_au 2000.

The postcode is remembered, so after the first run plain gridhour-au is enough. It is only used to pick your NEM region and is never sent anywhere. Western Australia and the Northern Territory are not part of the NEM, so their postcodes have no forecast.

gridhour-au --reset deletes the saved settings and the cache.

Usage

gridhour-au                          # your saved postcode or region
gridhour-au 3000                     # a new postcode (saved)
gridhour-au --region sa              # a region instead: nsw (Australian Capital Territory too), vic, qld, sa, tas (saved)
gridhour-au --mode green             # rank by carbon only (also: cheap, both)
gridhour-au --no-prices              # carbon only, skips the 5 MB PD7Day download
gridhour-au --at 2026-10-10T00:00Z   # start with the cursor at a given time
gridhour-au --theme slate --12h
gridhour-au --reset                  # delete the saved postcode, jobs, settings and the cache

Keys inside the app (? lists them all):

Settings and jobs live in ~/.config/gridhour-au/config.json.

Status bars and scripts

$ gridhour-au --line
⚡ 247g · 5c · green now

$ gridhour-au --line --best 3h
⚡ 247g · 5c · 3h best 11:00 (in 1d18h)

$ gridhour-au --tmux                 # same, with tmux colour codes for status-right
$ gridhour-au --watch                # same, updating in place
$ gridhour-au --json | jq '.jobs[] | {name, start: .best.from}'
$ gridhour-au --json --best 2h | jq '.best'
$ gridhour-au --ical > jobs.ics      # the best job windows as calendar events

If nothing in the next 48 hours is green, the line says greenest in 9h instead.

--json gives the current half hour, the next green slot, the best window for every job (its deadline and pieces included, with the carbon and price you would get starting now) and the full 48-hour series. Prices are in c/kWh and times in Coordinated Universal Time (UTC). It has "schema": 1; the JSON reference lists every field, its unit and when it can be null. Within schema 1, changes only add fields.

gridhour-au --json --best 2h adds a top-level best window for a two-hour job without touching the saved jobs: its UTC from and to, starts_in_minutes, mean carbon and price, and the parts it runs in. best is null if no complete window is available, and absent without --best. The ranking mode applies, as for saved jobs.

--json, --ical and --once exit with status 2 when there is no forecast at all (offline with an empty cache, or NEMWeb unreachable), so cron jobs and scripts can tell. They still print their usual output, errors included. --line and --tmux exit 0 regardless, so a status bar goes on showing the message.

For tmux, add set -g status-right '#(gridhour-au --tmux) ' to ~/.tmux.conf. For Waybar, Polybar or i3blocks, run gridhour-au --line every 300 seconds; cached runs don't fetch on every refresh.

Security and privacy

gridhour-au makes HTTPS GET requests to nemweb.com.au/Reports/ and nowhere else (redirects are refused), sends nothing about you, and has no runtime dependencies. Downloads are capped at 8 MB, and at 96 MB once unzipped. Every row is checked field by field before it is used or cached, and text from outside never reaches your terminal or tmux with escape sequences or format codes in it. Details: SECURITY.md.

Development

python3 -m venv .venv && . .venv/bin/activate
pip install --require-hashes -r requirements/dev.txt && pip install --no-deps -e .
pytest

The tests replay real NEMWeb reports from tests/fixtures/ (trimmed to the tables used); sockets are blocked while they run. More in CONTRIBUTING.md.

Licence

MIT, see LICENSE. Based on gridhour, also MIT.