v0.4.0 — Bloom plans your day

BloomCode v0.4.0 — Bloom plans your day

New

• Bloom plans each day. Every LeetCode problem has a rating: a zerotrac contest rating, or an estimate from difficulty, acceptance rate and tags. Each topic has a level that rises or falls with your results. Every attempt counts as Strong, OK or Struggled, judged from outcome, help, solve time, first-try acceptance and confidence. Bloom picks popular unseen problems near your level, weakest topics first, and gives a reason for each.
• Ideas come back, not the same problems. After a struggle you get easier practice in that topic. Otherwise, on the review date, you get a hidden Transfer check: a different, unseen problem with the same core idea, shown without its topic. Two solo solves two weeks apart retire an idea.
• LeetCode problem bank. Download every algorithm problem rated up to 2,000, premium included, from Settings. Daily plans no longer follow NeetCode 150 or Blind 75; those lists stay in the library.
• Plan changes in chat. Ask Bloom to add or swap problems; nothing changes until you confirm.
• Claude Code as a tutor provider. Settings → AI tutor runs Bloom on Codex or Claude Code, each on your own plan.
• Rating chips, per-topic training levels on Topics, a "Checks due" tile, and Too easy / Too hard and "Accepted first try" on finished attempts.

Install

Download BloomCode-darwin-arm64-0.4.0.zip, unzip, and replace BloomCode.app in Applications. This preview supports Apple silicon macOS. Existing study workspaces are retained. The first launch upgrades the database by adding tables only, so older builds can still open it.

After updating, open Settings → Problem bank and download it once so Bloom can plan from the whole LeetCode set.

The app is ad-hoc signed, not Developer ID signed or notarized. macOS may require approval in Privacy & Security for a downloaded app you trust.

Optional AI setup

The app includes Python workflow sources and its MCP adapter. Install Python 3.11+ and create the separate AI environment once. Re-run the pip install line after updating: this version adds the Claude Agent SDK.

python3 -m venv "$HOME/Library/Application Support/BloomCode/ai-runtime"
"$HOME/Library/Application Support/BloomCode/ai-runtime/bin/python" -m pip install -r /Applications/BloomCode.app/Contents/Resources/ai/python/requirements.txt

Restart BloomCode. Codex needs file-based sign-in (~/.codex/auth.json, or CODEX_HOME/auth.json); Claude Code needs claude auth login. Each uses the signed-in account's plan allowance, never an API key. Ordinary study features work without AI setup. No credentials or study data are included in the archive.

Known limitations

• Bloom's first plan of the day takes about 20–100 seconds. With the tutor off, the built-in rules plan instantly.
• Hidden transfer checks need the tutor; with it off, due ideas come back as visible topic practice.
• Estimated ratings are approximate (about ±150). Opening a problem in the library marks it as seen, so it can't become a later transfer check.
• Conversation history and coaching checkpoints are separate from study-database backups.
• The Python interpreter and its packages are not bundled; the one-time setup above is required for AI.

Validation

148 TypeScript tests, 64 offline Python tests, typecheck, lint, formatting, build and disposable runtime checks passed locally and in CI. Real Codex and Claude Code runs on a disposable database covered daily planning, a transfer check and its report, and confirmed chat plan changes. The packaged app was checked against a disposable database, including the problem-bank download and a Bloom plan on Codex.