README
LLM Daily Review — AI App Reviews from Hacker News
Automated daily testing & scoring of LLM apps posted on Hacker News. Find the best AI tools before everyone else — free, open source, no login required.
Live portal → tokenstree.eu
Every day at 15:00 UTC, this system scrapes Hacker News, finds LLM-related app submissions, tests each one in an isolated Docker container, and scores it across 11 criteria — publishing results with recommendation badges and a weekly Top 5 newsletter.
What it does
15:00 UTC daily
→ Scrape HN front page (top 30 items)
→ Filter: LLM / AI agent / generative AI tools only
→ Deduplicate against previously reviewed apps
→ Spin up Docker container per app (sandboxed, isolated)
→ Test: install, run, interact, benchmark
→ Score: 11 weighted criteria → normalized 0-100
→ Publish to tokenstree.eu with badges
→ Friday: Weekly Top 5 newsletter
Recommendation tiers
| Badge | Threshold | Meaning |
|---|---|---|
| ⭐ Strong candidate | Score ≥78 AND novelty ≥7 | Genuinely innovative, works well, worth adopting |
| 👀 Worth watching | Score ≥57 | Solid project with clear value |
| 🔍 Niche | Score 35–56 | Useful for a specific audience |
| ⏭ Skip | Score <35 or differentiation ≤3 | Too similar to existing tools, broken, or thin |
Scoring criteria (11 weighted, normalized to 100)
| Criterion | Weight | What we measure |
|---|---|---|
| HN Sentiment | 15% | Points, comments, positive/negative signals from community |
| Novelty | 11% | How original is the approach? Real differentiation vs wrappers |
| Current relevance | 11% | Fit with 2025-2026 AI landscape and use cases |
| Differentiation | 11% | What sets it apart from existing LLM tools? |
| Performance | 10% | Speed, reliability from automated benchmarks |
| Ease of use | 8% | QA results, UX quality, time-to-value |
| Ease of integration | 8% | API quality, SDK, plugin ecosystem |
| Documentation | 7% | README quality, examples, guides |
| Maturity | 7% | Commit history, open issues, maintenance signals |
| Community | 7% | GitHub stars, contributors, activity |
| System requirements | 5% | Setup ease, resource demands |
When a criterion is N/A, its weight is redistributed proportionally.
API
# Results for a specific date
GET https://tokenstree.eu/api/results?date=2026-04-14
# Calendar data (days with results)
GET https://tokenstree.eu/api/calendar?year=2026&month=4
# Weekly Top 5
GET https://tokenstree.eu/api/weekly-top5
GET https://tokenstree.eu/api/weekly-top5?week=2026-W15
Skills (open source scoring rubrics)
All LLM skills live in /skills/ — fork and improve them freely.
| Skill | Purpose |
|---|---|
hn-scraper | Fetch and parse HN front page |
app-identifier | Classify and filter LLM-related apps |
app-scorer | Score apps across 11 criteria |
app-benchmarker | Automated performance testing |
weekly-top5 | Friday ranking logic |
newsletter | Daily + weekly email generation |
Running locally
git clone https://github.com/vfalbor/llm-daily-review
cd llm-daily-review
npm install
cp .env.example .env # fill in: ANTHROPIC_API_KEY, SMTP credentials, GITHUB_TOKEN
node src/orchestrator/run.js # run a review cycle
node src/web/server.js # start the web UI on localhost:3000
To set up the daily cron (runs at 15:00 UTC):
bash cron/setup-cron.sh
Stack
- Runtime: Node.js 20 (ESM)
- Web: Express + SQLite (better-sqlite3)
- LLM scoring: Claude API (Anthropic)
- Testing: Docker (isolated containers per app)
- Email: Nodemailer
- Hosting: Linux VPS + nginx reverse proxy
Newsletter
Subscribe free at tokenstree.eu:
- Daily digest: every tested app with scores
- Weekly Top 5: Friday summary of the week's best LLM apps
Sent from info@tokenstree.com — check spam on first delivery.
License
CC BY 4.0 — fork, adapt, use commercially. Attribution required. Skills are intentionally designed to be community-adaptable. Fork and submit PRs.
Built with Claude Code · Free forever · tokenstree.eu
Contributors
2,893 commits
Similar repos
Languages
Python
92.0%
JavaScript
5.4%
HTML
2.5%
README
LLM Daily Review — AI App Reviews from Hacker News
Automated daily testing & scoring of LLM apps posted on Hacker News. Find the best AI tools before everyone else — free, open source, no login required.
Live portal → tokenstree.eu
Every day at 15:00 UTC, this system scrapes Hacker News, finds LLM-related app submissions, tests each one in an isolated Docker container, and scores it across 11 criteria — publishing results with recommendation badges and a weekly Top 5 newsletter.
What it does
15:00 UTC daily
→ Scrape HN front page (top 30 items)
→ Filter: LLM / AI agent / generative AI tools only
→ Deduplicate against previously reviewed apps
→ Spin up Docker container per app (sandboxed, isolated)
→ Test: install, run, interact, benchmark
→ Score: 11 weighted criteria → normalized 0-100
→ Publish to tokenstree.eu with badges
→ Friday: Weekly Top 5 newsletter
Recommendation tiers
| Badge | Threshold | Meaning |
|---|---|---|
| ⭐ Strong candidate | Score ≥78 AND novelty ≥7 | Genuinely innovative, works well, worth adopting |
| 👀 Worth watching | Score ≥57 | Solid project with clear value |
| 🔍 Niche | Score 35–56 | Useful for a specific audience |
| ⏭ Skip | Score <35 or differentiation ≤3 | Too similar to existing tools, broken, or thin |
Scoring criteria (11 weighted, normalized to 100)
| Criterion | Weight | What we measure |
|---|---|---|
| HN Sentiment | 15% | Points, comments, positive/negative signals from community |
| Novelty | 11% | How original is the approach? Real differentiation vs wrappers |
| Current relevance | 11% | Fit with 2025-2026 AI landscape and use cases |
| Differentiation | 11% | What sets it apart from existing LLM tools? |
| Performance | 10% | Speed, reliability from automated benchmarks |
| Ease of use | 8% | QA results, UX quality, time-to-value |
| Ease of integration | 8% | API quality, SDK, plugin ecosystem |
| Documentation | 7% | README quality, examples, guides |
| Maturity | 7% | Commit history, open issues, maintenance signals |
| Community | 7% | GitHub stars, contributors, activity |
| System requirements | 5% | Setup ease, resource demands |
When a criterion is N/A, its weight is redistributed proportionally.
API
# Results for a specific date
GET https://tokenstree.eu/api/results?date=2026-04-14
# Calendar data (days with results)
GET https://tokenstree.eu/api/calendar?year=2026&month=4
# Weekly Top 5
GET https://tokenstree.eu/api/weekly-top5
GET https://tokenstree.eu/api/weekly-top5?week=2026-W15
Skills (open source scoring rubrics)
All LLM skills live in /skills/ — fork and improve them freely.
| Skill | Purpose |
|---|---|
hn-scraper | Fetch and parse HN front page |
app-identifier | Classify and filter LLM-related apps |
app-scorer | Score apps across 11 criteria |
app-benchmarker | Automated performance testing |
weekly-top5 | Friday ranking logic |
newsletter | Daily + weekly email generation |
Running locally
git clone https://github.com/vfalbor/llm-daily-review
cd llm-daily-review
npm install
cp .env.example .env # fill in: ANTHROPIC_API_KEY, SMTP credentials, GITHUB_TOKEN
node src/orchestrator/run.js # run a review cycle
node src/web/server.js # start the web UI on localhost:3000
To set up the daily cron (runs at 15:00 UTC):
bash cron/setup-cron.sh
Stack
- Runtime: Node.js 20 (ESM)
- Web: Express + SQLite (better-sqlite3)
- LLM scoring: Claude API (Anthropic)
- Testing: Docker (isolated containers per app)
- Email: Nodemailer
- Hosting: Linux VPS + nginx reverse proxy
Newsletter
Subscribe free at tokenstree.eu:
- Daily digest: every tested app with scores
- Weekly Top 5: Friday summary of the week's best LLM apps
Sent from info@tokenstree.com — check spam on first delivery.
License
CC BY 4.0 — fork, adapt, use commercially. Attribution required. Skills are intentionally designed to be community-adaptable. Fork and submit PRs.
Built with Claude Code · Free forever · tokenstree.eu
Similar repos
Contributors
2,893 commits
Languages
Python
92.0%
JavaScript
5.4%
HTML
2.5%