English | 繁體中文
Turn messy Google Maps reviews into actionable, anti-hallucination intelligence for any AI Agent.
🌐 Official Website (tldreview.app)
- 🔒 Personal Dining Pocket List & Memory Vault (Beta): Save restaurants into 📌 Wishlist (
wishlist), ❤️ Liked (liked), or 💣 Avoid List (disliked) with custom tags and personal notes—and query saved spots near you using 0 SerpApi credits (see How to Use the Dining Vault below).
- 💎 Zero-Cost Metadata Extraction: Surfaces ⏳ Popular Times & Wait Time, 💰 Price Level & Live Open State, and 📋 Latest Menu Photos with zero extra credit usage.
- ⚔️ Multi-Venue Compare (
--compare) & 🗺️ Area Discovery (--top 3): Compare 2–3 venues side-by-side in a table, or scan the top 3 nearby spots, strip sponsored ads, and crown the 🏆 True Organic Winner. - 🗣️ Traveler's Point-and-Speak Card: Cross-language lookups (
ja,ko,th,fr, etc.) automatically preserve original local-language dish names in parentheses—e.g.,Braised Pork Belly (豚の角煮)—so you can point at your phone screen to order.
- 📍 "Nearby" Coordinate Search & 📸 Signboard/Menu Photo Lookup: Ask directly for "FamilyMart near me" or send a photo of a store signboard/menu (with built-in EXIF GPS extraction).
TL;DReview (https://tldreview.app/) is a zero-dependency Google Maps review intelligence engine built for AI Agents. It supports 6 intuitive workflows:
- 🔗 Paste any Google Maps URL (including short links & redirect wrappers).
- 📍 Ask for a "Nearby" Place (e.g., "FamilyMart near me"—locks onto the closest matching branch via GPS coordinates).
- 📸 Send a Photo of a Store Signboard or Menu (extracts store name, address, and EXIF GPS to jump straight into reviews).
- ⚔️ Compare Multiple Venues Side-by-Side (
--compare) (generates a comparison table of Must-Try, Red Flags, Wait Time, Price, and Best For). - 🗺️ Scan & Audit Top-N Nearby Places (
--top 3) (strips sponsored ads, audits fake reviews, and finds the true organic winner nearby). - 🔒 Personal Dining Pocket List Vault (
--vault-save/--vault-query, 🧪 Beta) (save places towishlist,liked, ordislikedand find nearby saved spots using 0 credits).
Google Maps' built-in summary is at most three generic lines. Meanwhile, pasting a raw Google Maps link directly into ChatGPT, Claude, or Grok fails due to JS anti-bot walls and LLM hallucinations:
| Stage | Pasting a Raw URL into a General AI | Using 🦊 TL;DReview |
|---|---|---|
| 1. Reading Reviews & Anti-Hallucination | Built-in AI web fetchers only see an empty JS shell, falling back to outdated blogs or hallucinating dishes out of thin air. | Extracts verified review text & rating distributions via API. 100% evidence-bound—sections without evidence are omitted. |
| 2. Signal-to-Noise Curation | Even copy-pasted reviews are flooded with one-word spam or outdated 5-year-old comments. | Filters out reviews >2 years old and ranks: ① Local Guide deep-dives ➔ ② Critical ★1–★3 reviews ➔ ③ High-like consensus. |
| 3. Fake 5-Star Radar | General AIs take 4.8★ ratings at face value and miss review-for-freebie promos. | Scans review text & bimodal 1★/5★ distributions to explicitly warn you about "5-star review for a free drink/dish" promos. |
| 4. International Travel Routing | Querying a restaurant in Japan, Korea, or Europe in English/Chinese misses 90% of local insights. | Infers local official language (40+ countries) to fetch native reviews (with English fallback) and builds a Point-and-Speak ordering card. |
| 5. Open Status & Menu Price Guardrails | Outdated review prices or walking to a store only to find it closed or cash-only. | Surfaces Price, Live Open Status, Popular Times queues, and Latest Menu Photos right in the header. |
Ever look up a great restaurant with your AI, only to forget what to order—or which places to avoid—a week later? Now you can ask your AI to save and organize restaurants for you:
Status (status) |
Meaning | What You Can Tell Your AI (Examples) |
|---|---|---|
📌 wishlist (Want to Go) |
Bookmarked to try later | "Save this place to my wishlist—note: take my parents here next time." |
❤️ liked (Went & Loved It) |
Visited and loved it; save your rating & notes | "I ate here today and loved it! Move it to my liked list, give it 4.8 stars, and note: ask for lighter broth." |
💣 disliked (Avoid List) |
Visited and disliked; never fall for it again | "I tried this place and it was terrible—put it on my disliked list." |
- ✨ Smart Status Upgrades (Zero Data Loss): When you visit a place that was on your 📌 Wishlist (
wishlist) and tell your AI "I went there and loved it (liked)!", TL;DReview automatically upgrades its status toliked, records yourvisited_atdate and personal rating, and preserves all previously saved must-try dishes, insider tips, and notes! - 🏷️ Custom Tags (
--tags): Tag any restaurant with custom labels likedate-night,family-friendly,late-night,ramen, ortaipei-daan.
Ask your AI anytime—queries run locally with GPS distance calculation and use 0 SerpApi credits:
- Find Wishlist Spots Nearby: "Is there anything on my wishlist near me right now?"
- Revisit Favorites by Tag: "Show me places I've
likedthat are taggedfamily-friendly." - Check Your Avoid List: "What restaurants are on my
dislikedblacklist?"
Because AI Agents often run inside ephemeral cloud VMs (like Meta Muse or Grok) where ~/.tldreview/sync.key can be wiped when a container resets, v3.0.0 includes Passphrase Escrow—inspired by iCloud Keychain Recovery, 1Password Secret Key, and Signal PIN:
- 🗣️ Recover with One Memorable Sentence (100% In-Chat, No Terminal Needed):
- Set Passphrase in Chat: Tell your AI "Help me set a sync passphrase", and your AI will provide an honest disclosure (stating that the phrase stays in chat history as a spare key to restore your vault) and ask for a memorable phrase (e.g. "Grandma makes the best braised beef noodles"). The AI runs
--vault-escrow --passphrase "..."in the background and never echoes the passphrase back in chat, only sharing the publicescrow_id. - One-Sentence Recovery on Any New Device/VM: On any new device or after a VM reset, simply tell that exact phrase to your new AI (which runs
--vault-recover --passphrase "...") to restore your sync key and dining pocket list! - Terminal Alternative for the Privacy-Conscious: If you prefer not to leave the passphrase in chat history, run
python3 core/tldreview_tool.py --vault-escrow --passphrase "<passphrase>"directly in your local terminal.
- Set Passphrase in Chat: Tell your AI "Help me set a sync passphrase", and your AI will provide an honest disclosure (stating that the phrase stays in chat history as a spare key to restore your vault) and ask for a memorable phrase (e.g. "Grandma makes the best braised beef noodles"). The AI runs
- 🛡️ No Silent Key Loss or "Parallel Universe" Vaults: Non-secret
~/.tldreview/vault_idis stored separately from~/.tldreview/sync.key(0600). Ifsync.keyis missing or corrupted whilevault_idexists, TL;DReview hard-errors and refuses to silently generate a second key, prompting you to recover via--vault-recoveror--vault-import. - 🔍 Honest Threat Model & Durability Contract:
- What E2EE Protects Against: Even the cloud relay server (Cloudflare Worker / KV) never sees what restaurants you saved. All encryption happens client-side before upload.
- Agent-as-Custodian Model: Because the AI Agent decrypts and queries your vault, the Agent's runtime holds the key (
~/.tldreview/sync.key) and caches plaintext locally at~/.tldreview/vault.json. The Agent is already the custodian; the recovery passphrase is a "spare key", not a secret to hide from the Agent. For a dining vault, chat history is a practical place to store a spare key (though this pattern must never be used for high-sensitivity secrets like banking passwords). - Honest Passphrase Tradeoffs: ① Weak passphrases can be brute-forced offline if the blob is obtained—mitigated via minimum strength checks, memory-hard
scryptKDF, and strict server-side rate limiting (max 5 escrow reads/min per IP); ② Forgetting your passphrase (without a key/JSON backup) means permanent data loss—there is no support desk that can reset it.
- 📝 Obsidian Note Export: Set
TLDREVIEW_OBSIDIAN_DIR=/path/to/obsidian/vaultto automatically export a clean Markdown card for every saved restaurant.
Note
🧪 Beta Notice: The Personal Dining Vault is currently in v3.0.0 Beta testing. Feedback and suggestions are welcome!
- Sign up at the SerpApi Dashboard to get your API Key (the free tier includes 250 searches/month, and querying your Personal Dining Vault uses 0 credits).
- Set
SERPAPI_API_KEYin your environment or connect your favorite AI Agent:
| Platform / Agent | Integration Method | Highlights | Guide |
|---|---|---|---|
| xAI Grok Bot | Cloud Computer Attachment | Persistent sandbox + Secret management | 👉 Grok Setup Guide |
| Meta Muse | Linux VM Sandbox | Sentinel network allowlist integration | 👉 Muse Setup Guide |
| Claude / Cursor | Model Context Protocol (MCP) | One-step mount into Claude Desktop / Cursor | 👉 MCP Setup Guide |
| LangChain / CrewAI | Python @tool Decorator |
Drop-in tools for multi-agent workflows | 👉 LangChain Guide |
| CLI / Local Terminal | Zero-Dependency Script | Markdown & JSON output with 24h cache | 👉 See CLI Usage Below |
🦊 TL;DReview Digest: Lao Zhang Beef Noodle House
📍 Address: Sec. 1, Xinsheng S. Rd, Da'an District, Taipei City
🏷️ Type: Noodle Shop | ⭐ 4.3 (Total: 1,820 reviews, Analyzed: 24 key reviews)
💰 Price: $200–400 | 🕒 Status: Open · Closes 9 PM
🛎️ Service & Options: Dine-in | Takeout | Cash only
---
🍜 Must-Try / Core Highlights
- Braised Half-Tendon Half-Beef Noodles: Melt-in-your-mouth tendon, thick shank slices, and rich herbal-beef broth (strong consensus).
- House Sichuan Pepper Beef Tallow: Adding half a spoonful dramatically elevates the aroma.
👍 Strengths & Pros
- Lightning-fast service; bowls arrive within 5 minutes even during peak hours.
👎 Weaknesses & Red Flags
- Broth leans salty; ask for extra clear broth if you prefer lighter flavors. Weak AC in summer.
💡 Operational Insights & Rules
- Cash only; NT$150 minimum spend per person. No signs of fake-review promotions.
🤫 Insider Tips
- Peak hours are Sat 12:00–14:00; arrive before 11:45 AM to skip the line. Choose wide hand-cut noodles.
🚗 Transit & Parking
- 6-minute walk from MRT Daan Park Station (Exit 1); park under the Jianguo Elevated Road lot.
---
💡 Want to save this place to your `wishlist` or mark it as `liked` / `disliked` in your Personal Dining Vault? Just let me know!
(See examples/sample_report.md for more examples.)
export SERPAPI_API_KEY="your_api_key_here"
# 1. Google Maps URL / "Nearby" GPS Search / Signboard or Menu Photo
python3 core/tldreview_tool.py "https://maps.app.goo.gl/sample" --lang en --markdown
python3 core/tldreview_tool.py "FamilyMart near me" --lat 25.0330 --lng 121.5654 --lang en --markdown
python3 core/tldreview_tool.py --image ./signboard.jpg --lang en --markdown
# 2. Multi-Venue Compare (--compare) / Area Discovery Top-3 Audit (--top 3)
python3 core/tldreview_tool.py "Lin Dong Fang Beef Noodle" "Yong Kang Beef Noodle" --compare --lang en --markdown
python3 core/tldreview_tool.py "ramen near me" --lat 25.0330 --lng 121.5654 --top 3 --lang en --markdown
# 3. Personal Dining Vault: Save to Wishlist (--vault-save)
python3 core/tldreview_tool.py --vault-save \
--name "Lao Zhang Beef Noodle" --lat 25.0330 --lng 121.5654 \
--status wishlist --tags "beef-noodle,daan,family-friendly" \
--must-try "Half-Tendon Half-Beef Noodles,Sichuan Pepper Beef Tallow" \
--user-notes "Take parents next time, ask for lighter broth" --markdown
# 4. Personal Dining Vault: Upgrade from 'wishlist' to 'liked' after visiting
python3 core/tldreview_tool.py --vault-save \
--name "Lao Zhang Beef Noodle" --status liked --user-rating 4.8 \
--user-notes "Parents loved the tendon noodles!" --markdown
# 5. Query Your Vault by Status, Tag, or GPS Radius (0 SerpApi credits!)
python3 core/tldreview_tool.py --vault-query --status wishlist --lat 25.0335 --lng 121.5640 --radius-km 3.0 --markdown
python3 core/tldreview_tool.py --vault-query --status liked --tag "family-friendly" --markdown
# 6. Passphrase Escrow (--vault-escrow) & One-Command Recovery (--vault-recover)
python3 core/tldreview_tool.py --vault-escrow --passphrase "Grandma makes the best braised beef noodles" --markdown
python3 core/tldreview_tool.py --vault-recover --passphrase "Grandma makes the best braised beef noodles" --markdown
# 7. Vault Diagnostics (--vault-status) & Portable JSON Backup (--vault-export / --vault-import)
python3 core/tldreview_tool.py --vault-status --markdown
python3 core/tldreview_tool.py --vault-export ./my_dining_backup.json
python3 core/tldreview_tool.py --vault-import ./my_dining_backup.json --markdown- Contribute an Adapter: Check out CONTRIBUTING.md and open a Pull Request!
- Buy Us a Coffee: If TL;DReview saves you time and helps you dodge tourist traps, consider supporting ongoing development:
👉 Buy Me a Coffee (https://buymeacoffee.com/dry4514)
Licensed under the MIT License.