Recode

A terminal-based spaced repetition tool for practicing any code from memory. Write implementations, compare against reference solutions with a side-by-side diff, get AI hints when stuck, and let the SM-2 algorithm schedule your reviews.

Features

• Spaced repetition — SM-2 algorithm schedules reviews so you spend time on what you actually need to practice
• Side-by-side diff — your attempt vs. the reference solution, line by line
• AI hints — Socratic nudges that guide you without just giving away the answer
• AI suggest fix — targeted bullet-point suggestions when you want more direct feedback
• OpenCode chat modal — conversational help in-context while you study a problem
• Agent-like chat tools — presets (/nudge, /test-me), TODO capture, diff split view, and /health
• 158 themes — full terminal.sexy palette, switchable live from the command palette
• Extensible — add any problem by dropping a .py, .jl, or .R file into your writable problems_dir

Screenshots

Problem list with SM-2 schedule	Side-by-side diff
SM-2 rating (Again / Hard / Good / Easy)	AI Socratic hint

AI suggest fix — targeted bullet-point suggestions

Requirements

• Python 3.10+
• uv (recommended), pipx, or pip
• A Gemini or OpenRouter API key
• OpenCode CLI on your PATH for in-app chat (auto-started by Recode)

Install

Homebrew

brew tap ever-oli/homebrew-tap
brew install ever-oli/homebrew-tap/recode

PyPI

uv tool install recode-cli
# or
pipx install recode-cli

Local dev

git clone https://github.com/ever-oli/recode.git
cd recode
uv sync
uv run python -m recode

Configuration

Recode reads environment variables from your shell, a local .env, or ~/.config/recode/.env. Use recode --paths to print the exact runtime directories for your machine.

On first run, Recode seeds its bundled problem set into your writable problems_dir so generated and imported problems live beside the defaults instead of inside the installed package.

Example .env:

# Option A — Google Gemini
GEMINI_API_KEY=your_gemini_api_key_here
AI_PROVIDER=gemini

# Option B — OpenRouter
OPENROUTER_API_KEY=your_openrouter_api_key_here
AI_PROVIDER=openrouter

# Optional: override the editor (default: hx)
EDITOR=nvim

# Optional: override runtime locations
# PROBLEMS_DIR=/path/to/your/problems
# DB_PATH=/path/to/recode.db

# Optional: OpenCode server URL for in-app chat modal
# OPENCODE_SERVER_URL=http://127.0.0.1:4096

# Optional: disable automatic `opencode serve` on first chat request
# OPENCODE_AUTOSTART=0

Run

Once installed:

recode

Useful non-interactive commands:

recode --version
recode --paths
recode --doctor

Chat modal behavior (OpenCode)

By default, pressing c in a problem will auto-start a local OpenCode server if it is not already running. The chat modal shows a live status line (connected, auto-started OpenCode, or offline). In chat, type /health to print OpenCode connection diagnostics. Use /help for chat commands and /clear to clear chat history in the modal. Chat also supports /hint, /fix, /todo ..., preset prompts, and /split (or Ctrl+D) to show diff + chat side by side.

If you prefer running it yourself (or disabled auto-start), start it manually:

opencode serve --port 4096

Usage

Inside a problem:

Workflow

1. Select a problem from the list and press Enter
2. Press e to open your editor and write the implementation from memory
3. Save and close the editor
4. Recode shows a side-by-side diff of your attempt vs. the reference
5. Use Hint or Suggest Fix if you need AI assistance
6. Rate your recall (Again / Hard / Good / Easy)
7. SM-2 schedules the next review automatically

Adding Problems

Problems are plain .py, .jl, or .R files. Drop them into the writable problems_dir from recode --paths and they will appear in the list on the next refresh (r). TensorPoly ships as a bundled collection and is copied into your writable problems directory on first run. The built-in importer currently supports Exercism Python and free LeetCode problems.

A problem file contains two things:

# The reference solution Recode compares against
SOLUTION = """
import numpy as np

def sigmoid(x):
    return 1 / (1 + np.exp(-x))
""".strip()

# The prompt shown in the problem list
DESCRIPTION = "Implement the sigmoid function using NumPy."

36 curated ML problems are included covering ResNet, Transformers, GANs, U-Net, ViT, and more.

License

MIT