Belle

@thedbeaudoin @thecamjwright You should try @Linkup_platform for deep research:)

more sophisticated than that, those aren’t great for deep research either tbh. we’ve tested and tested but have similar issues as Claygent. we do a combo of direct website scraping, LLMs, and 3rd party data tools for every trigger. and depending on the trigger the agent determines what combo of those & tools is needed.

Is it just me, or is Claygent weaker when the potential "sources" of that data are unknown? For example: if I ask Claygent (Argon) to determine if a company is using a certain technology it seems to frequently report false negatives. GPT, Claude seem to cast a much wider net.

@SatyaNaaksh Try out this prompt optimizer too, I've had great results with it prompt.linkup.so

So I’m adding a small 'Query Transformer' step right before Tavily. I pass the raw user input through a focused, lightweight prompt (can be iterative if needed) that cleans and reformulates it into an optimized search query. Then I feed that to Tavily. This gives me the best of both worlds: Sonar-like automatic query handling + Tavily’s full transparency and control.

I was happily using Perplexity Sonar as my main internet access tool for one of my stable projects. It worked great for months, gave clean results, good synthesis. Then one day they quietly raised the minimum API credits required. No announcement, no prior notice. Just higher billing suddenly.

@KTMudak Also try out @Linkup_platform's MCP. You can fetch whole sites but it's also great for deep research and more complex, multi-hop queries

Teach AI to research web pages properly Agents don't see websites the way you do. Most of the time they see page code and mostly plain text LLMs can't read the web out of the box. They need an MCP server for the task: > Firecrawl MCP – scraping + crawling a whole site > Jina Reader MCP – any URL to clean markdown, good for PDFs and academic papers > Tavily MCP – search + extraction from the current web, good at viewing pages > Exa MCP – AI-native semantic search: companies, code, research > Playwright MCP – when the site runs on JS and normal fetch doesn't work > Apify MCP – ready Actors for LinkedIn, Amazon, Instagram and other platforms > X MCP – Twitter/X through the official API, not HTML scraping All of them have free credits Personally I mostly use Playwright MCP. Not because it's better than everything else, I just got used to it and my tasks haven't needed more yet Links below

@zoepark_builds You should test on @Linkup_platform :)

the quiet race to own ai agent search is heating up tested tavily against tinyfish last week. tavily wins on speed but tinyfish finds niche forums my scraper missed. cost difference was $0.12 per 1000 queries. sometimes the cheap option is worse.

@zemse_ Even better, try @Linkup_platform :)

For having an agent do deep research, it needs a powerful search. Brave search api seems to be the only option out there, but I face a buggy paywall. Interestingly you can use duckduckgo (even if it has a captcha) github.com/zemse/ddg-rs but I really wish they officially start serving agents at this point

@jeremyphoward Linkup @Linkup_platform

What search providers are you all using with openclaw/pi/opencode/etc? Brave; serpapi; gemini; ...? Got any favorites?

@frangf @tavilyai @ChatGPTapp @claudeai @n8n_io GPT web search can be great for quick, lightweight stuff – but it’s a black box. You don’t get the same level of control you would with a dedicated provider like @Linkup_platform. It’s hard to tell what the model actually looked at – whether a full page or just a small snippet

Why should I pay @tavilyai if @ChatGPTapp and @claudeai already search the web? I know the answer is "for agents in @n8n_io". Fair. But I want real use cases: what have you built with it and what does it save you? #Tavily #n8n #AIagents

@ekpodar @StewartalsopIII @chainlank @premium Even better, try @Linkup_platform :)

@StewartalsopIII @chainlank @premium I think the model being used here doesn’t have access to current live information because it was likely trained before recent events To feed it live data, you can use SerpAPI, Tavily, Brave, or the Perplexity API. If it’s a live service, you may also want to optimize it for speed

hello @premium your bot is not working, i need my full data please

@mourginakis @usr_bin_roygbiv You should try @Linkup_platform, more cost effective and returns 2-3x more sources than other providers

@usr_bin_roygbiv Next unlock is finding a free or dead cheap search API to be able to do agentic research on massive amounts of text on the internet

getting WAY better results from qwen 27b on a 5090 than deepseek v3.2 from the api. Like 50% term bench, text classification/generation on some custom resume tailoring stuff, literally 3.5x faster on a 5090 w/o dflash or anything stock gguf model. Like sonnet 4.6/opus 4.5 #'s

@IamAdiG If you haven't checked out @Linkup_platform yet, their MCP server is great

first exa, now parallel good times for ai search

@KShivendu_ @championswimmer @badlogicgames @Kimi_Moonshot You should try @Linkup_platform

@championswimmer @badlogicgames @Kimi_Moonshot Which search api did you like the most?

been working with @Kimi_Moonshot K2.6 for the past 2 hours "accidentially" and ... it's really good!

@championswimmer @badlogicgames @Kimi_Moonshot Even better, you should try @Linkup_platform :)

@badlogicgames @Kimi_Moonshot The world knowledge of the open models out of Moonshot/Zai/Xiaomi have been a bit low though. What I've found useful is to always have a search MCP turned on (Exa/Brave/Perplexity), and tell model to search about things before planning. That truly gets it to Opus levels.

@miguelsaavedra @tavilyai @ExaAILabs @serp_api You should check out @Linkup_platform . It's easy to integrate and is the most affordable I've tried

Best search API? So far I’ve used @tavilyai, @ExaAILabs, @serp_api

@sdoi24592304 @AmeliaBarty I use @Linkup_platform

@AmeliaBarty Oh, I use perplexity API for search, any alternative recommendations?

Talk to any ex-Perplexity employee (there’s so many because of their ridiculously high turnover), and they’ll tell you the same thing: Aravind is *obsessed* with Twitter/X He literally spends 90% of his time thinking about how to get twitter engagement instead of running his company Supposedly this has been a key point of tension between him and cofounders Denis + Johnny, who effectively run the company in his stead The obsession sorta makes sense though, given that the only reason Perplexity exists is they happened to ride a viral twitter wave in 2023/24 and have been floundering ever since

@RyanWMark @kunchenguid Would love to see your benchmarks! Did you try it with you.com and linkup.so?

@kunchenguid Holy… I just force it to use dedicated apis for search: brave, exa, tavily, etc

alright agent nerds, if you care about your tokens and usage limits, pay attention to the tools you give to your agents. i built a benchmark that compared various browser tools for agents, and here's an example of their massive difference in cost and latency doing the same task

@OdinOnTheElba @DataChaz @jun_song I switched to Linkup's web search API and found it cleaner to integrate!

@DataChaz @jun_song I am running it in Hermes and really struggling to get it to search the web with my tavily API. Multiple tweaks of the model file and it's still leaking internal monologue into my output. Any suggestions?

🚨 `Super Gemma 4 26B Uncensored` is insane. @jun_song is COOKING AGAIN ♨️♨️♨️ he just dropped SuperGemma4-26B-Uncensored GGUF v2 and it is already trending on Hugging Face. This thing absolutely smokes the regular Gemma-4 26B. The specs: → 0/100 refusals. It is actually uncensored. → Fixed all the tool-call and tokenizer jank. → 90% faster prompt processing. → Sharper, smarter, way more capable responses. → It is the perfect local beast for llama.cpp. It runs on around 18-22 GB VRAM (the Q4_K_M file is 16.8 GB), meaning you can even run it on 16 GB GPUs. A 31B version is in the works and should be out soon. Pull this version on @huggingface below ↓

@1louder @TheAhmadOsman Even better, try @Linkup_platform :)

@TheAhmadOsman If you need high-quality search out of the box Brave/Tavily add clean indexing, relevance scoring, and LLM-ready structured summaries with free tiers. SearXNG is free, unlimited, and private but needs middleware to match cloud options.

Using local LLMs? Make sure to setup web search for them Tell your favorite agent to setup SearNg for you Give that to your local LLMs (tell an agent to set that up as well) Watch them become way more intelligent and efficient You're welcome

@SaranBluri @claudeai @Columbia @nyuniversity congrats!!

We won 3rd place + $100 in the Linkup track at the @claudeai Hackathon 2026 (@Columbia x @nyuniversity )!

@ArenaManCapital @MikeFritzell @smartkarma Or try @Linkup_platform they have a great google sheets integration:)

@MikeFritzell @smartkarma If you're feeling adventurous, a DIY solution could be to use AI with exa.ai or Braze Search API to get news and dump it in Excel, your own dashboard, or really anywhere you want!

Koyfin offers unlimited tickers in your watchlists, which you can then use to track news on the individual companies. Hard to beat at $39/month.

@JoshExile82 @CodingAcct @Teknium Even better, try @Linkup_platform :)

@CodingAcct @Teknium It was late so probably didn't type it out the best. Yes, right now there are lots of different providers like Brave, Firecrawl, etc. So was just curious which one people recommended to work the best with hermes and chatgpt 5.4 if I'm going to pay just want to use a good service.

So if I'm using Chatgpt 5.4 as my main model for Hermes @Teknium do we still have to use a seperate service for internet? For some reason, I thought using that model had internet baked in. I'm sure my old age is just getting to me lol