Tools

On 2026-06-20 @heyshrutimishra (reshared by Nous Research) tweeted “Hermes Agent is so good right now…,” signaling user enthusiasm about Hermes Agent’s performance.

BULLY DELUXE is a new release from artist @GoodAssSub, announced on X/Twitter on 2026-06-20.

AhaCreator claims a database of more than 5 million creators and 100,000+ registered users.

AhaCreator 3.0 is presented as an end-to-end AI agent for influencer marketing that automates creator discovery, vetting, campaign management, content review, and international payments.

The post highlights @dieworkwear's recent flag-themed merch as a rare example of quality in a trend the author labels 'slap a flag on it,' dated 2026-06-19.

The post identifies the knit as a Lemmermayer cardigan available from retailer Maus & Hoffman and gives fit guidance: opt for a slightly loose size so the sweater forms a gentle 'pooch' at the waist when fastened, and avoid making the garment suction-fit for a better look.

The post by @DeanTTraining promotes pairing Bulgarian split squats with the Rogers Squat Pro, framing the combo as an ideal setup for the exercise and asserting it is “the best way to do ’em.” The message is backed by a demonstration video and a linked clip from Wong (@WongEXE___) showing the Bulgarian split squat technique.

Understand Anything is an open-source plugin for agent platforms like Claude Code, Codex and OpenCode that analyzes repositories and builds an interactive knowledge base that explains the project rather than merely listing files.

Charles (@charles_irl) announces a co-release with Z Lab of six state-of-the-art DFlash speculators for Alibaba Qwen 3.x, touting a measured throughput exceeding 1,000 output transactions per second for Qwen 3.5 122B-A10B on a B200.

Luke Alonso posted an NVFP4 dump of GLM 5.2 sized at 467 GB.

GLM 5.2, according to @beaverd on 2026-06-20, runs locally at no cost and reliably follows complex, explicit multi-step instructions that the author says Opus could not execute.

@AlpacaAurelius tweeted on 2026-06-20 that customers are seeing "insane" hair growth after using their collagen.

Zeely AI announces that its avatars have been updated to display more natural emotions and on-camera reactions to avoid 'robotic' faces.

Dreamina Seedance 2.0 Mini is available in CapCut, emphasizing speed and cost efficiency: roughly double the generation speed and about 30% lower costs versus Seedance 2.0 Fast.

GLM 5.2 NVFP4 REAP 469B is running on a 3× DGX Spark cluster, delivering a 256,000-token context window with about 4.4 tokens/second throughput.

GLM-5.2-REAP is praised for proactively checking its work and catching errors before delivering results, a behavior the author finds highly valuable.

Author highlights that Z.

Kickbacks, introduced by Andrew McCalip on 2026-06-11, monetizes the Claude code-generation spinner by inserting auctioned ads during wait times.

Kyle E.

LiquidAI shows that overloading a model context with many tools increases latency and selection errors.

Liquid AI released two 350M-parameter retrieval models, LFM2.5-Embedding-350M and LFM2.5-ColBERT-350M, claiming end-to-end retrieval latency down to 1.5 ms and 'best-in-class' multilingual/cross-lingual performance across 11 languages.

The X post by @9rizumu dated 2026-06-18 provides a first look at a project called 'Reflections' Trainer, signaling an initial public preview.

AutoGTM, recommended by @hasantoxr, promises an automated GTM workflow: paste your product URL, let AI research your market in ~45 seconds, surface target companies, then generate and send personalized email outreach continuously.

@hamptonism describes the experience of an agent that pre-checks funding rates, whale movements, centralized exchange depth, and on-chain flows to give trading signals overnight.

GlobalGPT is presented as an all-in-one AI workspace (glbgpt.com) that aggregates over 100 models for $10/month.

Author @ginacostag_ announced on 2026-06-19 that GlobalGPT has made two large models—Gemini 3.5 Flash and Claude 4.5—public and free.

The post announces Lightricks' open-source LTX-2 project with a GitHub repo (github.com/Lightricks/LTX-2/), accompanying documentation (docs.ltx.video/open-source-m…), and a Hugging Face collection (huggingface.co/collections/L…).

LTX Trainer enables non-experts to train custom video models by speaking plain-language goals to Claude; an agentic skill then generates the training config and runs the training pipeline automatically.

The post by @DeanTTraining praises a Ninja Creami Deluxe–made treat as tasting and looking like a Dairy Queen item while offering substantially better nutrition: about one-third the Calories and five times the protein of the Dairy Queen option.

Trainer @DeanTTraining effusively praises a new Prime leg-extension machine posted 2026-06-19, calling it the greatest ever and lavishing hyperbolic praise on Prime’s design team.

Hermes Agent received a brief but emphatic endorsement from @DavidOndrej1 on X/Twitter on 2026-06-19, claiming it is "so good right now." The post signals high user enthusiasm or perceived improvement, though the author offered no supporting metrics, use cases, or comparisons to substantiate the claim.

Author @DataChaz, writing on 2026-06-19, says that using Microsoft Teams at major corporations a few years ago left them sympathetic to current users' struggles (accompanied by a laughing/sweat emoji).

Jeremy Howard, quoted by @daniel_mac8 on 2026-06-19, praises Zai_org’s GLM 5.2 as matching or exceeding Opus 4.8 and GPT 5.5 in quality.

GLM 5.2 is drawing strong acclaim: AndrewCurran notes a flurry of positive reactions over 48 hours (June 2026), while Jeremy Howard praises Zai_org's release as matching or exceeding Opus 4.8 and GPT 5.5.

Author @chddaniel contrasts three AI tools: ChatGPT is described as providing ideas and theoretical guidance, Claude Code is framed as a no-frills coder that produces executable code, and Shipper is portrayed as an end-to-end platform that builds and operates a business at a top-1% level, implying operational and managerial capabilities.

The post from @ManhoodDecoded on 2026-06-19 (Twitter/X) is titled "6 Supplements that make you LAST LONGER IN BED" and names creatine as the first supplement claimed to increase sexual lasting time.

T-Rex, from researchers at UC Berkeley, NVIDIA, and Stanford, integrates vision, language, and tactile inputs to enable real-time robotic responses to contact.

Brian Roemmele (post dated 2026-06-19) announces a new 'Psyops Identification Tool and a Universal Analysis AI Prompt' offered to his subscribers, claiming it upends prior analysis.

Brian Roemmele announced on 2026-06-19 a 'Psyops Identification Tool' and 'Universal Analysis AI Prompt' that he says leverages larger datasets and powerful AI to reveal previously undetectable patterns; he markets it to subscribers, urges journalists to use it ('Don’t do news without it'), and instructs users to ask 'Is this a PsyOp?' to reevaluate analyses and past reporting.

Brian Roemmele highlights a Psyops Identification Tool and Universal Analysis AI Prompt that an X subscriber reported ran on 53 historical news stories with striking results.

Rankhog, shipped by @anthonyriera and announced by the Ship or Die crew on 2026-06-19, was presented as a last‑minute launch — "5 days away from certain death." Crew reviewer @jackfriks lauded the product’s look and strategy, predicting "1 million MRR inbound" and calling the free‑audit CTA especially smart.

Rankhog is an AI Reddit growth product launched by Anthony Riera on 2026-06-19 that automatically constructs and iteratively improves a promotion strategy aimed at improving visibility on Google and in ChatGPT.

Alex Reibman, posting 2026-06-19 on X, says he already uses an agent-friendly email via @agentmail and is now looking for an agent-friendly bank account that supports real USD plus a debit card.

Alex_prompter posted a short solicitation on Twitter/X on 2026-06-19 asking whether readers have engineering teams that need AI enablement.

@alex_prompter urges users to "just run skills" to simplify Claude instead of juggling many features, promoting a paid Claude skills bundle that claims to turn Claude into more than 20 marketing and business specialists.

Exo is presented by @0xSero (19 June 2026) as a tool that removes manual tuning by automatically picking the optimal model, configuration, and harness for a user’s hardware based on local.ai benchmark results.

Andrew Feldman claims Cerebras delivers world-leading inference speed, arguing "fast tokens are the most valuable tokens." He says Google DeepMind's Gemma 4 achieves 1,500+ tokens/second on Cerebras — over 10× faster than GPUs and 15× faster than Claude Haiku 4.5 — and notes a private preview begins today with GA later this month for multimodal app builders.

Jeremy Howard (@jeremyphoward) reports that he has been using Fireworks AI (@FireworksAI_HQ) for model inference and suggests it outperforms other providers on speed.

The X post by @KookCapitalLLC (2026-06-17) endorses the @ether_fi crypto card, claiming it delivers the highest reward rate, instant ETH rewards, no requirement to lock tokens, and no in-app doxxing.

Kenji (@kenjidgn) posted on 2026-06-17 recommending the @TheKookLetter Telegram as an unmatched trading community for learning and active trading.

The post by @KookCapitalLLC/TheKookLetter claims $SYN jumped from an $8M market cap to $25M in roughly four days (about a 3x or +220% gain).

The post highlights OpenAIDevs' Codex Record & Replay (shared 2026-06-18), which captures a demonstrated, cross-application workflow once and turns it into an inspectable, editable skill.

Liquid AI released two new multilingual retrieval models—LFM2.5-Embedding-350M and LFM2.5-ColBERT-350M—claiming best-in-class multilingual and cross-lingual performance across 11 languages (including Arabic, Japanese, Korean, and major European languages) and end-to-end retrieval latency down to 1.5 ms; the post asserts Liquid AI is the only U.S.

Liquid AI announced two 350M-parameter retrieval models—LFM2.5-Embedding-350M and LFM2.5-ColBERT-350M—positioned for ultra-fast, accurate multilingual search.

Jitendra Malik highlighted that open datasets and models catalyzed breakthroughs in computer vision and suggested robotics could benefit similarly.

Bobyard is expanding its AI-powered takeoff and estimating platform, claiming top accuracy in the built world and rapidly opening new markets.

The author (@ivanfioravanti) will test Unsloth AI's GLM-5.2 the next day.

Hesamation highlights OpenAI Developers' new Codex Record & Replay feature as a major automation milestone, asserting it could replace most repetitive white‑collar work.

Author @hasantoxr reacted on 2026-06-18 to Fastlane (branded as Claude for Social Media by Gaurav/@gauravsbuilding), praising its ability to automatically create influencer personas, produce content,

Viktor launched in Microsoft Teams (approved and announced 2026-06-18) after running in Slack, positioning itself not as a reactive assistant but as a persistent AI employee that remembers past work, reads conversation context, and ships completed tasks into the team’s existing channel workflow.

Viktor, positioned as a chat-based assistant, is framed as “a proactive employee” by Matt Lis (CEO, CodeYourBrand) that learns company priorities and delivers without hesitation.

E-commerce is framed as a top business to start, with AI reportedly lowering barriers to entry.

The post by @EXM7777 (published on X/Twitter 2026-06-18) simply shares the domain machina.fyi with no additional commentary or context.

British Pullman, A Belmond Train is a short social post by @EvaLovesDesign published on Twitter/X on 2026-06-18.

The post highlights a tool that enforces AI access by citizenship, work authorization, and employment type while logging per-call spend to attribute costs.

el.cine (Twitter/@EHuanglu) shows on 2026-06-18 that the generative tool Seendance 2 can turn a Disney-style draft storyboard into a coherent, film-like commercial clip.

Tech commentator @duncancampbell applauds Midjourney’s release of the "Midjourney Scanner," calling it "truly amazing" and highlighting disruption in medical devices.

OrcaRouter warns that a 2025 breach of Microsoft 365 Copilot occurred when an AI read and obeyed a malicious email without any user click, illustrating a 2026 trend of agents being socially engineered.

OrcaRouter warns that the 2026 threat model is a 'haunted PDF' scenario where attackers in 2025 stole corporate data from Microsoft 365 Copilot by sending an email the AI read and obeyed.

Quander announced a $3M pre-seed led by Accel.

The post from @DataChaz highlights an open-source repository at github.com/bobek-balinek/cla…, gives a public shoutout to its creator @bobek_balinek for making the work available to the community, and asks followers to show support by starring the project on GitHub.

Poolside (@poolsideai) announced that open weights are now their default and released Laguna M.1, described as their most capable model, on 2026-06-18.

Claude Code introduced Artifacts on 2026-06-18: interactive pages created from a session (such as PR walkthroughs or living project dashboards) that teams can share through private links.

Brian Roemmele reports (ReadMultiplex, 2026-06-18) that Midjourney has launched a revolutionary full-body AI medical scanner designed for home use.

Unsloth AI released a locally runnable GLM‑5.2 2‑bit model that they say preserves ~82% accuracy after reducing size from 1.51 TB to 238 GB (‑84%), with GGUF weights and a setup guide available; it reportedly runs on a 256 GB Mac or similar RAM/VRAM systems.

Brian Roemmele posted on Twitter/X (2026-06-18) to announce a new subscriber-focused prompt titled “The Psyops Identification Tool and a Universal Analysis AI Prompt.” The brief message frames the content as a fresh tool or prompt being released to his subscribers, suggesting an intent to help identify psyops or provide a general-purpose AI analysis prompt.

Midjourney launched a full-body AI medical scanner on June 18, 2026, positioned as a consumer-facing tool to make preventive healthcare accessible by converting health monitoring into routine home use.

Kalshi announced (June 18, 2026) that prediction markets are launching in Canada via Wealthsimple, enabling millions of Canadian users to trade event contracts; the promotional post specifically cites a provocative example — Canadians betting on whether Canada will become the 51st U.S.

Benji Taylor highlights that his team leverages Grok Build to make bespoke internal design tools for producing high-quality videos.

Dreamina Seedance 2.0 Mini, now in CapCut, prioritizes speed and cost-efficiency: it produces videos about 2× faster than Seedance 2.0 Fast and costs roughly 30% less while keeping similar quality.

axanay (@MAD_as_25) built a local-first AI workflow engine (published 2026-06-17) that runs YAML-defined pipelines, summarizes files with Qwen3.5-2B via llama.cpp, and renders speech via Kokoro TTS; the implementation includes optional SQLite persistence, configured model paths, and a repo at github.com/Ananay28425/Kokor….

GLM-5.2 is now runnable locally per Unsloth AI (2026-06-18).

Jeremy Howard (@jeremyphoward) praised Zai_org’s GLM 5.2 on 2026-06-18, calling it a marvel comparable to Opus 4.8 and GPT 5.5.

swyx tweeted on 2026-06-18 that they asked the generative model @DevinAI to create a @tbpn-style breaking-news announcement card for the AIEWF speakers drop happening tomorrow; the model produced the

Author @kimmonismus, in the fourth post of a Twitter thread on June 17, 2026, urges readers to 'check it out' and links to tensordyne.ai.

El tuit de @guiomargdp (16 jun 2026) declara que reemplazará su "silla-armario" por el aparato mostrado en un video de Spacesthetic, llamándolo un "flechazo".

China unveiled an AI-powered, self-driving toilet robot for elderly and physically disabled users (announced 2026-06-17 on X by @interesting_aIl).

Zero automatically finds and integrates new tools for AI agents by searching 14,000 services, removing the frequent failure where agents stall when they need unfamiliar capabilities.

@hasantoxr argues that AI agents commonly stop when they hit tasks they lack pre-configured access for, forcing manual API/key/config setup.

Hassaan Raza (post dated 2026-06-17) frames Apple’s recent Siri preview against Apple’s 1987 'Knowledge Navigator' concept — an assistant that could see you, operate your computer, and look/sound human.

@hasantoxr reports they stopped paying Perplexity’s $20/month fee after discovering a GitHub repository that lets anyone build a personal AI engine on their own laptop.

Ollama, SearXNG and Vane are presented as three free open-source components for a privacy-first AI setup: Ollama runs models locally with no API keys, SearXNG provides private metasearch across Google, Bing, DuckDuckGo and 70+ sources without tracking, and Vane serves as the primary interface that integrates the two.

Ollama — a one-command installer that pulls a local model like Llama 3 onto a laptop, according to @hasantoxr (2026-06-17).

SearXNG provides private live-web access for on-device AI: @hasantoxr (2026-06-17) explains that while models can run locally, they need internet searches.

The author, @GoodAssSub, posted on 2026-06-17 asking for sources to buy authentic Ye merchandise originating in Türkiye.

Developer @godofprompt published a 60-second arcade game (video posted 2026-06-17) based on an "anti-AI-slop" banlist: players collect human-preferred words while dodging AI-overused terms.

Nitrosend: George Hartley argues that email dashboards ruled for 20 years and processed billions of emails, but that the dashboard itself became a bottleneck.

Gina Acosta presents a prompt that she says uncovers the most profitable, monetized problems in your niche and urges readers to save it.

FarzaTV built an interactive AI tutor powered by Claude Opus that performs pixel-accurate on-screen drawing and annotations (polygons, pointers) to visually guide users through complex tasks.

Kling 3.0 is promoted by @FynCas as a low-cost, ultra-fast tool for making trending “AI yapper” videos—about $1 each, produced in under a minute, and scalable instantly.