HTTP clients: evidence for the curl_cffi vs httpcloak matrix
Collected 19 Nov 2026. Four sources: what log runs today (from the code), what Scraping Enthusiasts said (log discord read of 7 channels, 4,100 messages), each library's own docs, and a live fingerprint test on this machine.
1. What log runs today
| Client | Platform | Library | TLS target | User agent | Needs |
|---|---|---|---|---|---|
| vendor/fbk | curl_cffi (Session + AsyncSession + WebSocket for MQTT/DGW realtime) | chrome136, FBK_IMPERSONATE, fallbacks 131/124/120 | Chrome 136 Windows hard-coded | REST, async, WebSocket | |
| vendor/twikit | X | curl_cffi AsyncSession | safari17_0 | Safari 17.5 macOS; guest client Chrome 122 | REST, async |
| vendor/instagrapi public transport | Instagram web | curl_cffi | chrome136 | Safari 11 / Chrome 136 | REST |
| vendor/instagrapi private API | Instagram app | requests (Android app) | n/a | Android app | REST, MQTT realtime |
| vendor/gabrielSlack | Slack | curl_cffi AsyncSession + ws_connect websocket | chrome150 | Chrome 150 macOS | REST, async, WebSocket |
| vendor/openLinkedinApi | curl_cffi Session | chrome150 | curl default | REST | |
| python/discordApi | Discord | curl_cffi Session (REST) + AsyncSession websocket gateway | chrome150 | captured from a real browser (Chrome 157) | REST, async, WebSocket |
| vendor/reverseChud | wreq-js (Node) | chrome_140 windows | wreq default | REST | |
| Camoufox | sign-in and browser runs, all platforms | Firefox 156 browser | Firefox | Firefox | cookies are made here |
All curl_cffi 0.16.3 built-in Chrome targets (chrome99 to chrome150) are labelled macOS in curl-cffi list. With no UA override, a chrome150 request says it is a Mac.
2. Live fingerprint test (tls.peet.ws/api/all, no proxy, this Windows machine)
Ground truth: real Google Chrome 157, Windows, launched through Playwright (channel chrome).
| Client and target | HTTP | JA4 | peetprint | Akamai H2 | User agent OS / version | Same as real Chrome? |
|---|---|---|---|---|---|---|
| real Chrome 157 | h2 | t13d1517h2_8daaf6152771_cb7bf5808d99 | fc97c1cdfb1409c9a9326c1b726d1dee | 52d84b11737d980aef856699f885ca86 | Windows 157 | (reference) |
| curl_cffi 0.16.3 chrome150 | h2 | t13d1516h2_8daaf6152771_806a8c22fdea | 67c3e9111bed9e7f03d2f21d6d88994b | 52d84b11737d980aef856699f885ca86 | macOS 150 | no |
| curl_cffi 0.16.3 chrome146 | h2 | t13d1516h2_8daaf6152771_d8a2da3f94cd | 1d4ffe9b0e34acac0bd883fa7f79d7b5 | 52d84b11737d980aef856699f885ca86 | macOS 146 | no |
| httpcloak chrome-150-windows | h2 | t13d1516h2_8daaf6152771_806a8c22fdea | 67c3e9111bed9e7f03d2f21d6d88994b | 52d84b11737d980aef856699f885ca86 | Windows 150 | no |
| httpcloak chrome-latest-windows | h2 | t13d1517h2_8daaf6152771_cb7bf5808d99 | fc97c1cdfb1409c9a9326c1b726d1dee | 52d84b11737d980aef856699f885ca86 | Windows 158 | JA4 + peetprint |
| httpcloak chrome-146-windows | h2 | t13d1516h2_8daaf6152771_d8a2da3f94cd | 1d4ffe9b0e34acac0bd883fa7f79d7b5 | 52d84b11737d980aef856699f885ca86 | Windows 146 | no |
| curl_cffi 0.17.0 chrome152 (Windows target) | h2 | t13d1517h2_8daaf6152771_cb7bf5808d99 | 1528e8ef528e7c8ec68fa406f9817b38 | Windows 152 | JA4 only |
- Akamai HTTP/2 hash was identical for every client: both libraries get HTTP/2 framing right.
- For the same Chrome version, curl_cffi and httpcloak sent byte-identical JA4 and peetprint (chrome150 vs chrome-150-windows; chrome146 vs chrome-146-windows). JA3 hashes differ run to run because Chrome permutes extension order.
- httpcloak's newest preset matched real Chrome 157 on JA4 and peetprint. curl_cffi 0.16.3 stops at chrome150 (JA4 one extension short of current Chrome). curl_cffi 0.17.0 adds chrome151/152 Windows and macOS targets and firefox153-155 Windows; its chrome152 matches current Chrome's JA4 but not its peetprint.
- curl_cffi 0.16.3 defaults to a macOS UA. httpcloak presets carry an OS (windows/linux/macos/android/ios) in the name and set UA, sec-ch-ua-platform and TCP/IP TTL to match.
- Header order was the same Chrome order on all clients.
3. Library docs
curl_cffi (lexiforest/curl_cffi, MIT; installed 0.16.3, latest 0.17.0)
- Python binding for a curl-impersonate fork. Impersonates browsers' TLS/JA3 and HTTP/2 fingerprints.
impersonate="chrome"tracks the newest; pin with a version, e.g.chrome124. - Sync
Sessionand asyncioAsyncSession(async with AsyncSession() as s: await s.get(...)). - WebSockets: "WebSocket handshakes go through your session, so they use the same browser impersonation (TLS/JA3 fingerprint), proxies, cookies and headers as your HTTP requests." Fragment reassembly, timeouts and retries, permessage-deflate.
- HTTP/3 with impersonation (
http_version="v3"). http and socks proxies. - Custom fingerprints:
ja3=,akamai=,extra_fp=. - Since 0.15.1,
curl-cffi updatefetches newer fingerprints from impersonate.pro without a new release; Safari, Chrome and Firefox updates are free, others are commercial. - Targets doc: "Chrome 151 and 152 have separate Windows targets (chrome151, chrome152) and macOS targets (chrome151_macos, chrome152_macos). The chrome alias selects the Windows profile. Firefox 153, 154 and 155 do not include dedicated HTTP/3 fingerprints." Suffix
a(e.g. chrome133a) = A/B-test variant. - Ecosystem: Scrapy integrations, requests and httpx adapters. A Node binding (
impers) exists.
httpcloak (sardanioss/httpcloak, MIT; Go core, PyPI/npm/NuGet bindings; tested 1.7.4)
- "Go HTTP client with browser-identical TLS/HTTP2 fingerprinting… JA3/JA4, Akamai fingerprint, header order. Supports HTTP/1.1, HTTP/2, HTTP/3, sessions, cookies, and proxies." Created 2025-12-28.
- Emulates TLS (JA3/JA4, GREASE, post-quantum X25519MLKEM768, ECH), HTTP/2 SETTINGS/WINDOW_UPDATE/priorities, QUIC parameters, HTTP/3 GREASE, the TCP/IP stack (TTL, MSS, window), Sec-Fetch and client-hint coherence, header order, cookie persistence.
- Presets per browser, version and OS, e.g. chrome-155-windows, chrome-158-macos, chrome-157-android, firefox-148-windows, safari-26-ios;
chrome-latest-*aliases track the newest shipped Chrome (158). describe_preset/load_preset_from_json: build any fingerprint from JSON, including from a tls.peet capture.- Connection lifecycle: Refresh (drop connections, keep tickets and cookies, like a tab reload), Warmup (load the page and subresources like a browser), Fork (parallel tabs sharing cookies and TLS state), session save and restore to JSON, HTTP/2 connection-age limit (5 minutes by default).
- Proxies: HTTP CONNECT, SOCKS5, SOCKS5 UDP ASSOCIATE and MASQUE for HTTP/3, runtime proxy switching, speculative TLS.
- Python API:
httpcloak.Session(preset=..., proxy=..., timeout=..., http_version=...),session.get/post/get_stream. The docs show no asyncio API for Python (Node and .NET have async methods). - WebSocket: the docs only mention that the WebSocket upgrade starts as an HTTP/1.1 request. There is no WebSocket client API in the docs for any binding.
- Native library prebuilt for Windows 10+, macOS 12+, Linux glibc; no musl.
4. What Scraping Enthusiasts said (every message on these clients, oldest first)
- 2024-11-13 #non-js Anthony: Hello, if you are in full request, some api (often paying) or tls fingerprint can help you to do it If you want to switch to headless browser, there are always paid solutions. If not, you can try to find libraries that can do it
- 2024-12-28 #non-js greggyNapalm: request will have another JA3/JA4 fingerprint
- 2024-12-28 #non-js AmericaN: even using tls_client or curlcffi
- 2025-02-24 #non-js ! TTG Boi: I'm trying to login into https://store.steampowered.com/ using cookies with requests in python, but it doesn't work. The cookie authentication only works when I use anti detect browsers like "Dolphin Anty" and manually add them. How can I achieve this in python using requests? (no selenium, pupeteer, etc.) I have tried to use curl_cffi but it doesn't work still.
- 2025-02-24 #non-js Nick Webson: try to copy a raw curl request from the browser, and then use verbose flag in curl_cffi and compare these requests looking for the diff
- 2025-03-06 #stealth-issues Nick Webson: that's the whole point, google trust you more if they see that you've been using this browser profile (cookies and other) for some time instead of just being a guy that respawned in the internet 5 minutes ago
- 2025-03-18 #non-js 0x676e67: It can only bypass protections such as TLS or HTTP2, and cannot do anything about JS challenges: https://github.com/0x676e67/rnet
- 2025-03-19 #non-js HDO BOX: Holy fuck!!! The site I want to scrape, which was using Cloudflare Turnstile, was giving me trouble! I tried everything to bypass Cloudflare, but nothing was working. However, this Rnet lib is working like a charm—no bot detection! btw, thank you so much for your cool lib!
- 2025-03-19 #non-js HDO BOX: I think the issue was with TLS, HTTP2, and a browser-like environment.
- 2025-03-19 #non-js gravi: Ohhh I didn't know about your work at all, it's really cool to have an alternative to TLS-Client but in Rust!
- 2025-03-19 #non-js HDO BOX: Since Rnet is for Python, are there any similar, modern, and up-to-date alternatives for Node.js?
- 2025-03-19 #non-js gravi: I can recommend this one: https://github.com/bogdanfinn/tls-client It’s in Go, but it exports a library that you can call with a C FFI client such as Koffi: https://koffi.dev
- 2025-04-16 #non-js casl22: ```py import random import time import datetime import string import tls_client def get_spotify_web_token(): session = tls_client.Session(client_identifier="chrome_105", debug=True) # session.proxies = {'http': 'http://127.0.0.1:8888', 'https': 'http://127.0.0.1:8888'} # Generate timestamps c_time = int(time.time() * 1000) s_time = c_time // 1000 totp = random.randint(479669, 879669) # Get current UTC date in YYYY-MM-DD current_date = datetime.datetime.utcnow().strftime("%Y-%m-%d") # Generate a random 7-character alphanumeric string random_suffix =
- 2025-06-06 #non-js Mozilla: https://github.com/Sahil1337/node-tls-client
- 2025-06-06 #non-js m0ltz: Good tip! Tried with
tls-client(https://bogdanfinn.gitbook.io/open-source-oasis/standalone-api-application) -- did not help. - 2025-08-21 #pextra-general rustace: how to make custom ja3
- 2025-08-21 #pextra-general 0xtem: if you use browser, there's no need to make a custom ja3, if not, use something like curl_cffi
- 2025-08-22 #pextra-general НОЖОЙ МОЙ ГЕРОЙ: It's not rly a bug, bet365 has some detection functions for detecting both playwrite and headless since avoiding playwrite headfull detection is quite easy (you need to spoof general playwrite fingerprints, bet365 detects "^__playwright" "^__pwInitScripts" "^__nightmare" "^__driver_evaluate" "^__webdriver_script_fn" "^selenium" "^selenium" "^chromedriver" "^puppeteer" "^cypress" "^cdc") props spoofing headless mode for bet365 and cloudflare quite tricky challenge there are most common detection techniques in the internet for chrome specifically but be sure bet365 and cf went way be
- 2025-09-05 #pextra-general gravi: Some alternative way to get what you want rather than relying on recommendations is to built and provide value to other people. They will naturally reach out to you to give you opportunities. I got most of my gigs helping around here and through GitHub comments / projects making me noticed and then people reaching through email. Of course you have to do something a bit different than the rest or do something technically interesting (I.e. something that comes to mind is @///captainbarnius with TLS-client in Go and @///0x676e67 with wreq in Rust/Python, and also the guy that created Camouflox)
- 2025-10-11 #pextra-general rustace: why random ja3 stiil be blocked ?
- 2025-10-11 #pextra-general Kar Sahatchian: Just another guess, maybe they have some kind of fingerprint database and are allowing only the known fingerprints (whitelist)? But that's very unusual I think, if I am not mistaken the vendors are usually implementing blocklists, because the ja3 fingerprints change frequently even between app versions
- 2025-10-12 #pextra-general Kar Sahatchian: Maybe you can try to replicate chrome or any other known browser? It's possible depending on your stack, on node.js the possible customization of the tls fingerprint is possible on several specific properties, meanwhile on python and go as I know if not fully then at least it's highly customizable
- 2025-11-07 #stealth-issues Kiko: Hello, I'm not sure if this is the right channel to ask, but I have a problem with anti-bot detection. I'm new to scraping and Python and decided to make a scraper so I could learn both, but I hit a roadblock: one of the sites I tried to scrape returns “Access Denied” when I launch it with the Selenium Chrome webdriver. I read that
curl_cffiis reportedly hard to detect, but I have no idea how to use it. If someone can help me with that site, I would be very grateful. I can provide code snippets and more information about the project. If someone can help, please DM me so we don't spam the - 2025-11-07 #beginner Kiko: Hello, I'm not sure if this is the right channel to ask, but I have a problem with anti-bot detection. I'm new to scraping and Python and decided to make a scraper so I could learn both, but I hit a roadblock: one of the sites I tried to scrape returns “Access Denied” when I launch it with the Selenium Chrome webdriver. I read that curl_cffi is reportedly hard to detect, but I have no idea how to use it. If someone can help me with that site, I would be very grateful. I can provide code snippets and more information about the project. If someone can help, please DM me so we don't spam the c
- 2025-11-07 #stealth-issues bosniankicks: ive written a few libs in differnet langs to help others start off holup
- 2025-11-08 #beginner JWriter20: One message isn't really spamming the channel, plus others can use this to learn from in future. curl_cffi is not complicated, you just use it as a drop in for the requests lib in python. Also recommend: https://github.com/daijro/hrequests (has features like built in stealth headless browsers as well). Basically both of these libs spoof your TLS fingerprint (look up ja3 or ja4+, basically you can track request senders by looking at the packet metadata headers) and request headers so that the request looks legit, and it looks like a different machine is sending the request each time you make a
- 2025-11-12 #beginner marin: Can I control over 2000 Sessions of curl_cffi on a single fastapi project? ram and cpu resources are not a problem, im geting hit with Connection closed abruptly and TLS issues. anyone want to help me tackle this?
- 2025-11-24 #beginner Aytac: Hi everyone, so, there is a website, that i need to scrape some datas, plus i needed some click functionalities, so i used the Selenium for this. Everything is going perfect. But after some requests, I am getting blocked by the website ( because of the too many requests i assume ). let's say I have 500 links to scrape, but it gets blocked at the 30. I tried free proxies for rotation, even bought not free ones and tried, but none of them were working for the proxy changing in the Selenium. I am dead end stuck and couldn't find a solution for this blocking thing anymore. Should i use Seleni
- 2025-11-29 #programming kavislavis: wanna do async? asyncio, haven't seen any painful event loop bugs like the endless uncaught errors in JS or missed stacktraces in kotlin http? curl_cffi, no way any other client can outperform good ol' curl types? type hints are great. i've built pretty advanced stuff using pydantic and powerful py enums. validation? Pydantic also saves you captchas? opencv and keras available and well supported, go try solve the dd slider without calling py (I tried, don't recommend. it's possible tho) cdp? plethora of options, I find custom ws client + chrome-devtools-protocol lib best for it scale, rate lim
- 2025-11-29 #beginner ostrich: I have a website link that i need to get the jump on as soon as its content updates. right now its most likely not accessed by many people at all unless they had the same idea and info that I have. It's a direct link and isn't an indexed site + there is nothing on the domain of the site which points to this link so it's only really accessed by people via typing the URL into the browser. I can't find much online about this usecase but is there anything special I should know about this? I essentially need to refresh the site for like 12 hours until the contents of the site updates. right now I
- 2025-11-29 #beginner JWriter20: Ah ok got it, nvm feroxbuster is not the solution then. Unless the site you are targeting has something like a webhook, you will have to do a polling-based approach similar to what you're doing now. Maybe just throw it in a cron job that runs every 10 seconds with curl_cffi or something until the page updates and can be parsed
- 2025-12-28 #stealth-issues Luma: Try using a TLS client: ```py from curl_cffi import requests s = requests.Session() response = requests.get('https://downdetector.com/', headers={ 'accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,/;q=0.8,application/signed-exchange;v=b3;q=0.7', 'accept-language': 'en-US,en;q=0.9', 'cache-control': 'max-age=0', 'dnt': '1', 'if-modified-since': 'Sun, 28 Dec 2025 05:46:03 GMT', 'priority': 'u=0, i', 'sec-ch-ua': '"Google Chrome";v="143", "Chromium";v="143", "Not A(Brand";v="24"', 'sec-ch-ua-mobile': '?0', 'se
- 2026-02-04 #non-js saksham: I agree with this approach, I used jsdom to learn stuff, managed to bypass medium level security(though heavy work handling by httpcloak tbh) but yeah really useful for learning and understanding how it all happens.
- 2026-02-28 #beginner gravi: Maybe start by looking at TLS fingerprinting? Then try reverse engineering a simple antibot.
- 2026-03-23 #beginner axp: has anyone had any experience bypassing Imperva? Running into their WAF for a scrape and having issues trying to fetch cookies and pass them along to curlcffi. i cant figure out a long-term solution - its very strange, for the first few requests i get http 200, but then after a couple of times i get http500. I'm barely hitting them - its like 3 request over the span of like 2 minutes and then blocked, not enough to trigger any kind of rate limit. Does Imperva have a block list of datacenter IPs? My request headers are all correct so I'm guessing they block datacenter IPs? Don't have experience
- 2026-04-16 #non-js SwagiWagi: Wrote up how we tracked down a silent production failure to the TLS handshake itself - connections completing fine, valid responses, then garbage on every subsequent request in the session Turns out our JA3 was flagged. wrote up the full investigation: capturing the ClientHello with mitmproxy, what actually makes a fingerprint detectable (it's not just the ciphers, it's GREASE and padding too), and how curl_cffi fixed it at scale without the overhead of a full browser engine Also covers JA4 and why the move to sorted extensions changes the spoofing game https://dev.to/romclerix/how-we-rever
- 2026-04-28 #beginner Pooom: I've been reading about the basics of Go, and I'm creating a project using Surf. I'm gonna try to impersonate the browser with it.
- 2026-04-28 #beginner Pooom: If anyone here has experience with Surf or wreq... What do you guys think are the ods of bypassing tumblr antibot system with them ? should be enough ? (apart of the need for proxies...)
- 2026-05-18 #beginner gravi: I don’t know Surf but I’ve been really happy about wreq.
- 2026-05-18 #stealth-issues Stubbtoe: i get this for linkeidn. they are usually session bound within an ifarme, meaning you cant prot site keys in for success - they JA3/4 your set up and asign cookie ids to your fingerprint and also dont allow porting in, its all erver side rendered
- 2026-07-03 #non-js Akintouch: > yeah the http/1.1 thing is what's actually killing you. akamai wants http/2 and your proxy's dropping it down to 1.1, so you're flagged before your fingerprint even gets looked at. you want a proxy that connect-tunnels so the browser keeps its own tls + http/2 all the way through, not one that mitms and rebuilds the traffic. > > also patchright just patches the automation stuff, it doesn't touch the tls/http2 or engine fingerprint that akamai's actually reading. something like fortress (https://github.com/tiliondev/fortress) is recompiled chromium so the http/2 + ja3/ja4 come out right on t
- 2026-07-03 #non-js Arham: yeah fair, most "stealth browser" repos right now are exactly that, a stealth.min.js reskin with a fresh readme. the skepticism's earned. but on this one the copy-paste read doesn't hold up. any real chromium emits the correct ja3/ja4/h2 for free, so fortress never needed to touch tls. that part's on how it was pitched, not the tool. it's a fingerprint + raw-cdp thing, not a network-stack thing. on the proxy: a clean CONNECT tunnel keeps h2 end to end, agreed. so a 1.1 error under a chrome UA means the proxy isn't tunneling cleanly, it's terminating tls or not advertising h2 in alpn. easy to
- 2026-07-03 #non-js Akintouch: > also patchright just patches the automation stuff, it doesn't touch the tls/http2 or engine fingerprint that akamai's actually reading. something like fortress (https://github.com/tiliondev/fortress) is recompiled chromium so the http/2 + ja3/ja4 come out right on their own idk, that's what sound like. > on the proxy: a clean CONNECT tunnel keeps h2 end to end, agreed. so a 1.1 error under a chrome UA means the proxy isn't tunneling cleanly, it's terminating tls or not advertising h2 in alpn. easy to confirm, run tls.peet.ws / http2.pro through it. not h2 = there's your block. <:wtf:776752
- 2026-07-04 #non-js HistoriFy: as akin said in the original message - "all words made sense - but its completely false" same thing applied to his this section as well?
on the proxy: a clean CONNECT tunnel keeps h2 end to end, agreed. so a 1.1 error under a chrome UA means the proxy isn't tunneling cleanly, it's terminating tls or not advertising h2 in alpn. easy to confirm, run tls.peet.ws / http2.pro through it. not h2 = there's your block.??? so many things wrong here > so a 1.1 error under a chrome UA means the proxy isn't tunneling cleanly this doesn't mean that the proxy is always bad can also be origin - 2026-07-04 #non-js Akintouch: I didnt wanted to argue seing > on the proxy: a clean CONNECT tunnel keeps h2 end to end, agreed. so a 1.1 error under a chrome UA means the proxy isn't tunneling cleanly, it's terminating tls or not advertising h2 in alpn. easy to confirm, run tls.peet.ws / http2.pro through it. not h2 = there's your block. My point is not really about bashing, that fine if ppl dont understand how those stuff works, just dont come with an expert posture and authoritative argument And to reframe this how a proxy works (a regular, not a mitm) and CONNECT is for init the client <-> proxy tunnel, there are
- 2026-07-14 #proxy-related JWriter20: Speaking of which, you seen any good new ones? Here are a few from awhile back, but haven't been seeing a ton of new stuff https://github.com/tiliondev/fortress, https://github.com/sardanioss/httpcloak, https://github.com/CloakHQ/CloakBrowser
- 2026-07-14 #proxy-related DrSelenium: Haven't tried fortress or httpcloak, but CloakBrowser has weak stealth. Easily detected on https://pixelscan.net/fingerprint-check
- 2026-07-28 #beginner JWriter20: Probably just scrape from sephora or amazon, although scraping might not even be needed, it looks like sephora may have a product api, if there is an API, you should basically always use that first. If there isn't one that fits your use case, try a TLS scraper first on some of the key endpoints (I recommend httpCloak). if that doesn't work you probably need browser rendering, in which case you should use something like headless camoufox or patchright. To find the endpoints you should be scraping, go to the page with the information you want, right click, click inspect element, go to the networ
- 2026-08-28 #stealth-issues Ivan: hey, question about webgl spoofing on a server boxI run chrome under xvfb on a linux vps and patch UNMASKED_VENDOR/RENDERER to look like a normal windows machine, but the actual rendering is still swiftshader (ANGLE Google Vulkan). getting ~40% blocked by f5/shape on a gov sitealready ruled out the IP side — ran the same curl_cffi request through the same mobile proxy from a laptop and from the vps, identical responses byte for byte, zero blocks. so it's the browser not the networkis patching just the strings basically useless if they compare actual render output? does a real gpu on the box ma
- 2026-09-13 #pextra-general unknown: no, it's for phones with internet (but no gps) to be able to geolocate without gps
- 2026-10-22 #scraping-general Xetera: I mean I have its internet connection blocked so it doesn't make much of a difference either way
- 2026-10-25 #beginner JWriter20: C# was made by microsoft, which immediately makes me lose faith in it. Honestly I haven't given it a fair shake though, it could be a great language. Problem is all of the web automation tooling is either TS/JS, python, rust or go. I mean, especially if you're just scraping, you're gonna have like a 100 line script that can capture everything max, would likely only be like 20 lines. Also first before you use a browser, try tools like https://github.com/sardanioss/httpcloak or whatever equivalents are in C#. You almost always want to go for non-browser solutions if you can. Honestly I just re
- 2026-10-25 #beginner Fessus: I am usually both scrapping and processing as needed in a single app... Thanks for the httpcloak, I didn't know it is a thing and it sounds very useful.
- 2026-10-25 #beginner Krokzter: My stack is in C# as well, you can definitely do web scraping large scale with just C#. HttpCloak has C# bindings
- 2026-11-01 #programming protobluf: also is curlium basically curl-impersonate?
- 2026-11-01 #programming Akintouch: for human interaction, it's different from what you can find in js, it's a system trained on real user interaction (it's not AI, just mathlab statistical model) that reproduces real input and lives within the browser and is calibrated on fps (user input/fps are linked) so it replicates with high precision the correct frequency/velocity (it's not driven via cdp input) and for curlium, the origin is from curl impersonate yes. The main diffs are mostly internals (tls session share/resume, proxy masque/udp handling, UTF-8 normalization, and the FFI is pretty different for performance/allocation)
- 2026-11-06 #beginner Rabies Ronnie: F12 in browser -> find the request that holds relevant data -> try to repeat it with curl -> if curl succeeds, use preferred library to request it in a script, if curl fails verify that all the headers, etc are correct, if curl still fails use something like curl-cffi, if that fails, piggyback on a browser
- 2026-11-13 #scraping-general the owner (log's own account): Ive been using curl_cffi mainly for mocks but if anyone knows of anything better lmk
- 2026-11-13 #scraping-general JWriter20: I like httpcloak by <@1131722747760869409>