Your App Store reviews, read for you. Every Monday, as a GitHub Issue.
English | 简体中文
Fork this template, fill in your app id, add your own LLM API key — and every week an AI reads your app's newest reviews across every country's App Store, clusters them into complaint themes, crashes, feature requests, and reviews worth replying to, translates them into your language, and delivers the briefing as a GitHub Issue.
Zero servers. Zero database. Zero subscription. Runs entirely on GitHub Actions in your own repo, with your own API key. Commercial tools charge ~$199/month for this; here the only cost is your LLM usage (typically a few cents to ~$1 per week).
Real output (from a run against TikTok's public reviews, 237 new reviews across 3 storefronts):
237 new reviews across 3 storefronts · average 2.5★
Unjustified bans, strikes & age restrictions (~30, US/GB/DE) — Automated moderation banning users who claim no wrongdoing; opaque, bot-driven appeals that repeat the same reply.
"Automated moderation warned me for no reason (I posted the German national flag) and this led to a full account ban; my appeal was also rejected." [DE]
- Black screen / logo hang on launch, then crash (~20, US/GB/DE) — Started after a recent update. Heavily concentrated on iPhone 7 / 7 Plus and iPads on iOS 15.8 / 15.8.8 …
- "Black screen on launch since the update, iPhone 6s iOS 15.8.8... I need this for back to school." [US ★5] — Acknowledge the iOS 15.8.x/older-device crash; give timeline or workaround.
Every section is grounded in the actual reviews — counts are derived from the data, quotes are real (translated, with the origin storefront tagged), and empty sections say "None this week."
Plus a free dashboard on GitHub Pages — rating trends per storefront and every past digest as a web page:
You need: a GitHub account, an app on the App Store (yours or anyone's), and an LLM API key.
Click Use this template → Create a new repository (top right). Private is fine. You get an independent copy that you fully own — not a fork.
Edit reviewdigest.yaml in your new repo (the pencil icon on github.com works — no need to clone):
apps:
- id: 1234567890 # ← the number from your App Store URL:
# https://apps.apple.com/us/app/<name>/id1234567890
countries: [us, gb, de, jp] # or `major` (top 20) or `all` (~170)
language: English # digest language — review quotes get translated into itPrefer clicking to typing? Use the setup wizard — search your app by name, tick storefronts, copy the generated file:
In your repo: Settings → Secrets and variables → Actions → New repository secret
- Name:
ANTHROPIC_API_KEY, value: your key from console.anthropic.com - Or
OPENAI_API_KEYwithllm.provider: openaiin the config — any OpenAI-compatible endpoint works viallm.base_url
The key lives encrypted in your repo; usage bills to your account. No key? It still works — you get the stats plus a raw review listing instead of the AI analysis.
Actions tab → Review digest → Run workflow. A couple of minutes later your first digest appears under Issues, labeled review-digest.
Settings → Pages → Source: "GitHub Actions". From the next run on, https://<you>.github.io/<repo>/ serves your rating-trend charts and digest archive. Skipping this is fine — the weekly Issue is unaffected.
Every Monday 08:00 UTC (edit the cron in .github/workflows/digest.yml to taste), a new Issue arrives. Read it in 5 minutes, know what your users are angry about, what crashed, and who deserves a reply. Discuss in the Issue, link fixes, @ your team.
All options live in reviewdigest.yaml, documented inline. Highlights:
| Key | Default | Notes |
|---|---|---|
apps |
— | one or more App Store app ids; names auto-resolve |
countries |
[us] |
storefront codes, major (top 20), or all (~170) |
language |
English |
digest output language; quotes translated |
lookback_days |
8 |
window scanned per run (overlap is deduped) |
llm.provider |
anthropic |
anthropic or openai (any compatible endpoint via base_url) |
llm.model |
claude-opus-4-8 |
claude-sonnet-5 is the budget option |
output.type |
github-issue |
or file / stdout |
pip install -r requirements.txt
python -m reviewdigest --dry-run # full digest to stdout (needs LLM key in env)
python -m reviewdigest --dry-run --no-llm # free: stats + raw review listing
python -m reviewdigest.site # build the dashboard into _site/Flags: --force (digest even with 0 new reviews), --config path, --output stdout|file|github-issue, -v.
- Fetch — reviews come from Apple's first-party App Store web API (the same one apps.apple.com uses), newest-first, per storefront. No Apple credentials needed. Store ratings come from the official iTunes lookup API.
- Dedup — seen review ids live in state/ and get committed back by the workflow. A review is never reported twice. No database.
- Analyze — new reviews go to the LLM with a prompt tuned for operator-grade honesty: counts from the data, real quotes only, no invented numbers.
- Deliver — one Issue per digest; a markdown copy lands in digests/, which feeds the Pages dashboard. Weeks with zero new reviews are skipped.
Is this official? The review endpoint is Apple's own web API but undocumented. If Apple changes it, the fetch layer is isolated in reviewdigest/fetch.py and easy to swap (App Store Connect API is the documented fallback for your own apps).
How many reviews can it see? Up to max_pages_per_storefront × 20 per storefront per run (default 100). Very high-volume apps should raise it or run more often.
Android? iOS only for now. Google Play's API requires developer verification and a service account — PRs welcome.
Can I watch competitors? Yes — any public app id works. Reading what users complain about in competing apps is a feature list waiting to happen.
Privacy? Everything runs in your repo with your keys. Review text goes to your chosen LLM provider and nowhere else.
Email/Telegram/Slack outputs, Google Play, trend improvements — see CONTRIBUTING.md. The output layer is a single function; the fetch layer is one file.