The Learning Curve

A coding agent that respects students.

LearningCode shows every token it spends, plans before it touches your code, and runs on the phone you already own. Free CLI — powered by a Spark token plan.

macOS · Linux · Windows · Android (Termux)

See the cost

Every turn shows its tokens and price in the footer.

Plan before it edits

Read first. Nothing changes until you approve.

Use your phone

Pick up any session from a phone or Termux.

Plans from RM20/month

Subscribe to Spark

Quota

Every token, on screen.

Every turn spends quota and takes a seat — so the footer shows the price instead of hiding it. The only agent that tells a student what things cost.

Weekly budget meter

50% · 125k / 250k

Unlimited and no-cap states are shown honestly, never as /0K.

Session spend

sess 3k ≈$0.03

Priced per turn from the model catalogue. No rates? Tokens alone, never a made-up price.

Plan mode

Read first. Touch later.

Shift+Tab flips between research-only planning and full building. The model can read everything, but nothing changes until you approve.

  1. 1

    Plan

    Research only. Writers are hidden, and edits are refused with a reason.

  2. 2

    Approve

    Read the numbered steps, then /approve when you agree.

  3. 3

    Build

    Full access returns. Steps tick off as they finish.

What Plan mode allows

  • ✓ Read files, search, list folders
  • ✓ Read-only commands like git status and ls
  • ✕ Edit or write files
  • ✕ Install packages, commit, or delete

Refused actions say why, so you know exactly what to approve.

Helpers

Big jobs, a fresh mind.

Reading a dozen files to answer one small question fills the assistant's memory and makes it vaguer as it goes. A helper is a fresh copy of the assistant with its own clean memory — it does the reading, answers in a few lines, and forgets the rest.

Assistant its memory fills up scout clean memory reviewer clean memory planner clean memory worker clean memory one job each, at once a few lines back

One assistant. Every helper is a fresh copy with clean memory — it reads, works, and answers in a few lines, so the assistant's own memory stays short.

Four helpers ship with it

  • scoutlook around and report what's where
  • reviewera second opinion on work so far
  • plannerturn a goal into ordered steps
  • workerdo one small, complete piece of work

What you see

  • Which helper ran, and a one-line result — Ctrl+O opens the full answer
  • Several helpers at once, or one helper's answer chained into the next
  • Anything a helper changes is covered by /undo, like the assistant's own edits

Helpers need a model that can run more than one answer at a time, so they stay off on TLC-Spark. Switch with /model to use them.

Remote

Start on the laptop. Finish on the phone.

Run learningcode remote --resume in your project folder and scan the code. The agent keeps running on your computer, with its files, shell and Git. The phone picks up the same conversation.

Providers

Spark by default. Your own keys when you want them.

Spark is the built-in model. LearningCode also runs on the providers below, set up with learningcode --login. Switch models any time with /model.

model: TLC-Spark

Keys stay on your computer. Pricing for each model is read from its provider's catalogue, and the footer shows what the session costs.

First run

A tour, not a manual.

A 5-minute guided tour with its token price disclosed upfront. Type skip anytime, replay with /intro, opt out forever with --no-intro.

Keep working from your phone

Pick up any session from a phone or Termux. Plan, approve and track your checklist without a laptop.
plan · approve · quota · todos

Never stuck on an error

Every failure says what went wrong and the next step, in plain words.
doctor --fix repairs the common ones.

Download

One line, every machine.

Linux & macOS
curl -fsSL https://raw.githubusercontent.com/stemtrooper/learningcode/main/install.sh | bash
Windows (PowerShell)
npm install -g @stemtrooper/learningcode

Then run learningcode --login once, and learningcode starts the tour. Source and releases: github.com/stemtrooper/learningcode

Token plans

The CLI is free. The model runs on Spark.

Pick a token plan, paste it into learningcode --login, and every turn draws from your weekly budget — always visible in the footer.

Basic

RM20/month

  • 250k tokens per week
  • Oyen model (27B)
Subscribe

Plus

RM50/month

  • 1M tokens per week
  • Oyen (27B) and Rimau (70B) models
Subscribe

Quotas reset weekly. Full terms on spark.learning.com.my.

Momentum

Shipped, not promised.

0.14.1

Helpers, now with names

0.13.0

Starts sooner on a slow network

0.12.0

Your phone can run the commands too

0.11.0

Strict plan mode, for training wheels

Open source

Read the code. Open an issue.

LearningCode is MIT-licensed. The CLI, the tour, the plan mode and the changelog all live on GitHub, along with the issue tracker.

Start building in five minutes.

A token plan for the model. A one-line install for the CLI.