Guides & tutorials
Operator-tested tutorials on scraping, Apify, and AI social listening — from someone who runs 30+ live actors.
Centaline API: Ask Claude to Compare Hong Kong Property Prices via Apify MCP
Centaline publishes no official API. Wire Apify's official MCP server into Claude Code or Claude Desktop instead, and you can ask in plain English how Taikoo Shing's current asking prices compare with its recent deals — Claude calls centaline-scraper itself and does the math. We ran it for real: 33 transactions against 40 listings, for about $0.17.
2026-10-12 How to Use Jev for Social Listening: API Request, 7-Step Recipe and the Questions That Worked
A practical guide to using TypeSafe's Jev API for sentiment, relevance and intent labels on forum posts: the HTTP request shape, a 7-step recipe with the measured evidence behind each step, the full question definitions we used, cost per 100,000 posts, rate limits and pitfalls.
2026-10-12 Jev vs Claude Haiku: An Independent Sentiment Benchmark on 405 Cantonese, Thai and Singlish Forum Posts
We ran TypeSafe's Jev (jev-1.13.0) and Claude Haiku 5.5 on the same sentiment, relevance and intent questions over 405 real forum posts from LIHKG, Pantip and HardwareZone. Haiku led by 3 to 5 points on 150 held-out Cantonese posts; Jev cost less than half as much, answered in about 0.36 seconds, and its confidence scores were well calibrated.
2026-10-12 Cantonese Sentiment Analysis with LLMs: What Is Hard and What Accuracy to Expect
Why Cantonese forum sentiment is hard (slang, sarcasm, quoted replies, bad news as negative, brand as scene), and the accuracy two current AI models reached on 150 held-out LIHKG posts, with the definitions to settle before you measure.
2026-10-12 Jev Pricing: What It Costs per Post, per 100k and per Million (Measured)
TypeSafe's Jev costs $0.042 per million input tokens with free output. This page explains why tokens per request vary, shows our measured cost per post for six request shapes, compares it with Claude Haiku 5.5, and works through labelling a million forum posts a month.
2026-10-12 What Is Jev AI? TypeSafe's System One Model Explained, With What We Measured
Jev is TypeSafe's System One model: it returns typed answers with probabilities instead of generating text. This guide explains Choice, Score and Noul questions, how Jev differs from an LLM, its documented weak spots, and what we saw when we tested it on real forum posts.
2026-10-12 Jev vs OpenAI Decisions API: Accuracy, Cost and Speed on 266 Cantonese and Thai Forum Posts
We ran TypeSafe's Jev (jev-1.13.0) and OpenAI's Decisions API (gpt-6-luna) on the same sentiment, relevance and intent questions over 150 held-out Cantonese posts and 116 Thai posts. Sentiment was a tie, Jev led on Cantonese relevance, intent split by language, and Jev cost about 44% as much per post. OpenAI's relevance confidence was the more overconfident of the two.
2026-10-12 Thai Sentiment Analysis with LLMs: How Accurate Are Three Models on 116 Pantip Posts?
We tested Jev, OpenAI's Decisions API and Claude Haiku 5.5 on 116 real Pantip posts about six Thai brands. All three beat the trivial baselines on sentiment and intent (82.8% to 86.2% sentiment against a 72.4% always-neutral baseline), Haiku was best on intent, Jev was cheapest, and relevance scores were uninformative because 94% of the posts were relevant.
2026-10-11 Social Listening Thailand: Search Pantip Only in English and You Miss 4 in 10 Topics
Pantip is the 11th most-visited website in Thailand, and the first on that list that is not a global platform, a marketplace, a messaging app or an AI assistant. On 11 October 2026 we searched it for six brands, once by their Latin name and once in Thai script: the Latin-only searches surfaced 60 of the 104 distinct topics, and for four of the six brands the two spellings shared no topic at all. This guide covers how to pull Pantip topics and comments with the pantip-scraper-en Apify actor, why its search is not a live feed, when to expand a thread, and what a daily monitor costs.
2026-10-10 LIHKG Scraper Has No Official Docs: Why Thread List, Search and Full-Thread Are 3 Different Builds, Not 1
LIHKG has never published developer API docs, so anyone building a lihkg scraper has to reverse-engineer its api_v2 endpoint from scratch. The catch: its three read modes — board thread list, site-wide keyword search, and full-thread-with-replies — have different inputs, different output shapes and different pagination rules. This guide breaks down what each mode actually returns, where a DIY build gets stuck, and when a maintained actor is worth paying for instead.
2026-10-10 Apify Free Tier: How Many Scraper Runs Does $5 a Month Buy?
Most developer tools — AI APIs, automation platforms, hosting — bill a fixed amount every month whether you use them or not, so it's easy to assume Apify works the same way. It doesn't: the free plan costs $0, needs no credit card, and gives $5 of usage credit a month, billed only when an Actor actually returns data. This piece runs the numbers on ptt-scraper to show how many PTT posts $5 actually buys.
2026-10-01 Farming Webshare Accounts for Free Residential IPs vs. Paying for Apify Residential Proxy
An open-source tool called PETANIPROXY auto-creates Webshare accounts to farm free residential IPs for scraping. Its own README shows the captcha it solves is Webshare's signup verification, not your scrape target's anti-bot defense. This piece separates that mix-up from the real cost math: farming free IPs vs. paying for Apify residential proxy, and when a captcha solver is actually worth adding.
2026-09-26 Build vs Buy: The Real Cost of DIY Social Listening for Taiwan, Hong Kong and Singapore Forums
Taiwan's OpView publishes a standard-plan price of NT$39,800 (about US$1,250) a month, a one-year minimum contract, and a 3-month lookback window. If you just need brand sentiment across Chinese-language and Asian forums, ready-made Apify actors run pay-per-event from a few dollars a month and the raw data is yours. The real cost math on both sides, worked out.
2026-09-20 PTT Monitoring: Hire Someone to Watch PTT and Mobile01, or Let a Scraper Do Most of the Job?
A real Taiwan job posting pays NT$30,000–40,000/month (about US$1,100–1,480 once you add employer overhead) to have someone watch PTT, Dcard, and Mobile01 for brand chatter. The board-scanning slice of that job runs on a schedule with ptt-scraper and mobile01-scraper for about US$34/month — but copywriting, planning, and judgment calls still need a person. A duty-by-duty breakdown with real cost math.
2026-09-19 Monitor Taiwan's Bahamut Gaming Forum in English with Claude Code + Apify MCP
Global game teams shipping into Taiwan can't read Bahamut, its largest Chinese-language gaming forum — but patch reactions are sitting right there. Wire Apify's bahamut-scraper into Claude Code over MCP, ask in English, and the agent reads the threads and summarizes sentiment back to you.
2026-09-18 Hong Kong's CCL Index Is Up 12% in 2026 — Sha Tin First City Still Sold at a HK$1M Loss
Hong Kong's Centa-City Leading Index (CCL) hit 162.02 points in September 2026, up 12.43% year-to-date and its second-highest level in about three years — but Sha Tin First City and Tai Po Grand Deer both saw units resell at a loss in that same stretch. The CCL is a citywide weighted average across roughly 100 large estates, updated once a week; it can't tell you what the estate you're actually watching is doing. This guide shows how to pull estate-level transactions, live listings, and the CCL's own historical series yourself with the centaline-scraper Apify actor.
2026-09-18 Hong Kong's HOS Swap Math: One Buyer Pays HK$10,000/Month, Another Nets HK$2M — Check Your Target Estate's Real Price First
Hong Kong social media is full of Home Ownership Scheme (HOS) swap stories — one poster regrets paying 'over HK$10,000 a month,' another financed a Green Form HOS flat in Sheung Shui with a 95% mortgage and HK$10,000 monthly payments. Neither number applies to the estate you're actually considering. The government extended HOS/Green Form mortgage guarantee periods to 50 years in March 2024, but estate prices (Shan Lai Court: HK$2.99M-$3.38M) and management fees (HK$5.4/sq ft at Ki Wah Court vs HK$1.6/sq ft at Yue Nga Court) vary by estate. This guide shows how to pull real asking prices and transaction history for your target estate yourself with the centaline-scraper Apify actor.
2026-09-18 Verify Hong Kong Property Listing Claims Yourself: Is That Flat Actually For Sale? Check It Against Real Agency Data
Hong Kong's Anti-Deception Coordination Centre and Estate Agents Authority both warn against trusting social media property posts at face value. Centaline's own chairman has said only about 6.25% of secondhand listings carry a genuine mandate, and an agency survey found 73% of buyers hit listings still posted after the unit had sold. This guide shows how to cross-check deal_type, posting date and asking price yourself with the Centaline, 28Hse and Midland Apify actors, instead of trusting one social post.
2026-09-18 AI Deepfakes Impersonated Two Hong Kong Celebrities to Sell Stock Tips — Build a Public Evidence Timeline Before You Report It
Hong Kong feng shui master Su Man Fung and 900,000-subscriber YouTuber Ming Zai were both hit by AI voice-and-face deepfakes pushing stock tips — and their platform complaints got nothing but auto-replies. This guide shows brand-protection teams, journalists, and diaspora researchers how to use LIHKG-scraper's search mode to build a dated, linkable timeline of the public discussion before filing a formal report.
2026-09-18 Sushiro Apologized Same-Day, a Pop Star Responded Next-Day: A Fast Official Reply Doesn't Mean the Story Has Cooled — Track the Timeline with LIHKG-scraper
Sushiro apologized in the comments the same day its conveyor-belt video went viral, and singer Chen Xiaodong publicly responded the day after a baseball team cancelled his appearance — both official reactions came fast, but a news headline is only one snapshot in time. This guide shows brand and PR teams how to schedule LIHKG-scraper's search mode and compare reply_count / last_reply_at across runs to see whether a crisis discussion is still climbing or has actually gone quiet.
2026-09-18 LIHKG Search as an Early-Warning Net: Catch Complaint Threads at 10 Replies, Before They Reach the News Desk
Plenty of Hong Kong complaint threads start with a dozen replies in a finance or workplace sub-forum, long before they reach a news-worthy sub-forum. This guide shows brand-safety, compliance, and PR teams how to use LIHKG-scraper's search mode — which is site-wide by design, not sub-forum-scoped — to sweep many keywords at once (brand names, estate names, company names) and catch a signal before it becomes a headline, and diff thread_id lists across runs to spot what is new.
2026-09-18 Threads Keyword Search API: 3 Ways to Scrape Posts by Keyword — the Official API's Real Limits, Your Own Scraper, or an Apify Actor
"Threads has no search API" is outdated — keyword_search exists, but needs App Review, caps at 2,200 queries/24h, and skips like/repost counts. Compare it against a hand-rolled scraper and a ready-made Apify actor ($0.005/post, no login, full engagement data) to pick what fits your budget.
2026-09-14 Blocked by Cloudflare? First Figure Out If It's a Block or a JS Challenge — Proxies Don't Fix Both
A scraper that suddenly can't get past a Cloudflare-protected site looks like one problem, but it's usually two: a clean 403 with no verification step, or a 'Just a moment' JS challenge that evaluates your browser fingerprint. Cloudflare's own docs split these into different actions with different fixes — this piece works through both, then contrasts them with a real Block-tier case: LIHKG-scraper's default proxy exists to dodge a 429 rate limit, not a JS challenge.
2026-09-10 Free GitHub Proxy Lists vs Apify Residential Proxy: Which One Survives a Scheduled Scraper?
GitHub proxy-list repos like proxifly rack up thousands of stars and re-validate every few minutes — but a public list is public to everyone, including the IP-reputation databases anti-bot systems buy from. This piece uses a comparable open-source proxy list's own verified-vs-scraped ratio, Apify's official pricing, and one actor's default-proxy warning to lay out free lists vs Apify datacenter vs Apify residential proxy for anyone running a scheduled scraper.
2026-09-10 EDMW Forum Monitoring: HardwareZone's Front Page Turns Over Every 6 Minutes — How to Monitor Singapore's Biggest Forum Without Gaps
Singapore's loudest public conversation happens on a XenForo forum, not on a platform with an API. On 10 September 2026 at 12:49 SGT, all 20 non-pinned threads on the front page of HardwareZone's Eat-Drink-Man-Woman board had been replied to within the previous 6.4 minutes, and that single board's pagination runs to 32,456 pages. A once-a-day scrape of page one sees almost none of it. This guide shows how to pull the board index and thread replies yourself with the hardwarezone-scraper-en Apify actor, which boards carry commercial signal, and why replies — not listings — are what actually costs money.
2026-09-01 Build a no-code forum sentiment pipeline with Apify + n8n + GPT
Wire an Apify forum scraper into n8n, let GPT tag every post positive/negative/neutral, and get a Slack alert when sentiment turns — on a schedule, without writing code. Worked example uses Taiwan's PTT and Mobile01, but the same pattern fits any forum Apify covers.
2026-09-01 Build a no-code price-drop alert with Apify + n8n for any e-commerce site
Most e-commerce sites don't ship an official price-drop alert, and price-comparison sites won't always track the exact SKU you care about. This walkthrough uses Apify's Taiwan e-commerce actors (PChome and Ruten) as the worked example, wired into n8n's Compare Datasets node, to catch a price drop and fire a Slack alert — no crawler code, and the same shape works for any e-commerce site Apify covers, from Singapore grocery SKUs to anything else with a search-mode scraper.