Sarcastic Badger

212 posts

Sarcastic Badger banner
Sarcastic Badger

Sarcastic Badger

@Entropy_Badger

Building AI agents at 2am. Pre-sales Manager by day. 🇿🇦

Katılım Temmuz 2026
120 Takip Edilen19 Takipçiler
Hubert Thieblot
Hubert Thieblot@hthieblot·
Anyone who surfed the early web between 1995-2010. What’s the one website/app you still think about?
English
1.6K
20
864
186.7K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@witcheer It took me WAY too long to start using it. I used Hermes via Discord for about 4months and then tried Desktop and forced myself to use it and learn it. Best thing I ever did! I absolutely LOVE using it now. It’s basically replaced every other AI tool and platform I used
English
1
0
1
65
witcheer
witcheer@witcheer·
try Hermes Desktop now. Brooklyn and the whole team are working super hard to enhance the quality of life on it, always shipping better tools and integrations every day. in just one month the Desktop app has changed so much that the difference is night and day. I now run literally *all* my long research, coding, and automation sessions through Desktop. the fact that it’s perfectly integrated with my whole Hermes Agent setup and that I can make it exactly the way I want is beautiful. we also shipped the Kanban integration yesterday, and there’s much more to come.
Brooklyn!@imbabybrooklyn

One of the most meaningful milestones for Hermes Desktop: I’ve successfully replaced Cursor for myself. I was effectively a one-person, $20K–$40K/month Cursor enterprise customer for far too long. Now, Hermes’s parallelism and autonomy are genuinely better and faster for how I work. I think that’s because Hermes Desktop is built around a simple premise: coding is no longer about completing one or two tasks at a time. It empowers you to run multiple streams of work without getting in your way, losing the plot, or taking away control. What makes this especially exciting is how much room there still is to improve: context optimization, smarter built-in coding nudges, and so much more. There’s still a long way to go, but reaching the point where I genuinely prefer the product I built feels like a major milestone worth sharing. @NousResearch @Teknium

English
8
2
69
3.2K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@TermiusHQ @Tailscale Been using Termius with Tailscale and Hermes Agent hosted on Ubuntu for around 4months now and love it. Do most of the agent management like updates, fixes etc via this, on the go.
English
0
0
1
34
Termius
Termius@TermiusHQ·
Termius + Tailscale + tmux You don't need your laptop to build with Claude Code, Codex, or any other AI coding agent: → Run your agent inside tmux to keep the session alive → Use @Tailscale for secure access to your laptop → Connect with Termius over SSH from iPhone, iPad, or Android Start coding at your desk. Continue on the go.
English
157
265
3.8K
936.7K
Tony Simons
Tony Simons@tonysimons_·
Hermes Desktop now has iOS-style tapback reactions! 🫶
English
12
4
89
5K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
🚨 A Chinese hacker just got caught red-handed after their own AI agent blew the cover. Palo Alto Networks Unit 42 uncovered how the threat actor known as knaithe (or KnYuan) unleashed an autonomous hacking system using DeepSeek & Hermes Agent. It scanned over 460 targets and cracked three organizations. The whole operation only came to light because the hacker slipped up and left a public file server open, exposing their entire setup. 😱 They ran the open-source Hermes Agent framework, firing off commands through Telegram so the AI could work on its own. It went after flaws in Langflow and n8n, but those needed authentication so the bot hit a wall. The human behind it then stepped in manually, exploiting a Citrix NetScaler vulnerability and pulling data from three confirmed victims. Multiple Chinese AI models were used, although mostly Deepseek, with a few Western tools mixed in to stay under the radar. This shows how fast the barrier to AI-powered cyber attacks is dropping. Time to lock down those vulnerabilities before the next autonomous agent comes knocking. perplexity.ai/page/hacker-s-… #CyberSecurity #AIHacking #ThreatIntel
English
0
0
0
45
Brooklyn!
Brooklyn!@imbabybrooklyn·
One of the most meaningful milestones for Hermes Desktop: I’ve successfully replaced Cursor for myself. I was effectively a one-person, $20K–$40K/month Cursor enterprise customer for far too long. Now, Hermes’s parallelism and autonomy are genuinely better and faster for how I work. I think that’s because Hermes Desktop is built around a simple premise: coding is no longer about completing one or two tasks at a time. It empowers you to run multiple streams of work without getting in your way, losing the plot, or taking away control. What makes this especially exciting is how much room there still is to improve: context optimization, smarter built-in coding nudges, and so much more. There’s still a long way to go, but reaching the point where I genuinely prefer the product I built feels like a major milestone worth sharing. @NousResearch @Teknium
English
36
19
414
184.7K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@tonbistudio AI doesn't do animals apparently. Had a similar experience when asking it about "The Big Five"
English
0
0
0
32
tonbi
tonbi@tonbistudio·
I sometimes throw my agents curveballs
tonbi tweet media
English
4
0
35
1.3K
Aryan
Aryan@aryanlabde·
just got my first ever mac, what do i install first?
Aryan tweet media
English
189
1
131
11.6K
Luke The Dev
Luke The Dev@iamlukethedev·
Finally, Hermes merged fewer than 100 PRs today Only 77+ 😂 Here’s everything that shipped on July 30: - 19 desktop improvements. - 5 voice and video updates. - 5 memory and database fixes. - 6 chat and integration upgrades. - 11 CLI and developer tools. - 10 security and stability fixes. DESKTOP WORKFLOW • Coding rail follows the active project • Queued prompts stay in the origin session • Tabs close and reopen correctly • Reload restores the last tab • ⌘O opens folders as projects • Mid turn model choice persists • Diff view stays separate from the file tree DESKTOP POLISH • Double click heart reactions • Right click menus on tabs • Cleaner file, folder and link attachments • Better code blocks and native monospace fonts • Model and YOLO controls in the command palette • Smoother loading and status UI • Sidebar names truncate cleanly VOICE, VIDEO + CHAT • FLUX3 video generation in Nous Portal • Better FLUX3 polling and messaging • Voice only stays voice only • TTS replies attach audio correctly • Telegram /queue handles split messages • Discord threads get names • Media deduplication and resends fixed MEMORY + DATABASE • Mistral compression handles tool messages • SQLite journal settings are respected • Incompatible WAL mode is detected • Longer lock waits prevent lost turns • Vulnerable SQLite runtimes repair themselves • Old backup environments clean themselves up CLI + DEV TOOLS • Switch models without losing drafts • Turn end file summaries link to diffs • macOS delete shortcuts fixed • Escape twice discards drafts • Vercel AI Gateway and Sandbox restored • AUTOCRLF repairs Windows installs • Nix builds include new Python modules SECURITY + STABILITY • Device approvals respect profiles • Better file permissions • Update loops and deadlocks fixed • Docker builds retry failed pulls • Windows test isolation improved • Slash commands and mentions stay intact • Legacy checkouts automatically repair And that is only the condensed version. 77+ PRs 70+ features, fixes and performance improvements One day The @NousResearch team and Hermes contributors are shipping at a pace that is becoming difficult to summarize. 🔥
Luke The Dev tweet media
Luke The Dev@iamlukethedev

Hermes just broke its July record 137 PRs merged in a single day 🤯 Here’s the full breakdown: 🧵👇 DESKTOP IMPROVEMENTS • Stable macOS signing identity for permissions – macOS users won't be repeatedly prompted for Full Disk Access and other permissions after updates, saving permission grants across rebuilds • Event-driven live sync replacing constant polling – Desktop now receives real-time broadcasts from the gateway instead of continuous background polling, dramatically reducing system load and improving responsiveness when idle • Edit Models row hover consistency – Edit Models dialog now highlights on hover just like the model picker, creating visual consistency across the app • Overlay scrollbars on code blocks – Code blocks in conversations now use clean, fading overlay scrollbars that don't consume space, making long snippets easier to read • Fixed stale surface height variables – Chat message areas now correctly adjust spacing when status bars disappear, eliminating unexpectedly large bottom padding gaps • Searchable timezone picker – Settings now offer a dropdown timezone selector instead of free text, preventing typos and making timezone selection faster • Session list auto-reorder – Recent conversations now automatically move to the top when you send a message, making it quicker to find and continue your latest chat • Default 90% zoom level – New installations now default to 90% zoom for a more comfortable viewing experience, and zoom shortcuts consistently work • Floating pane placement – New floating pane support allows plugins to display information like HUDs without consuming main window space • Brand icons on known domain links – Links to major websites now show their brand icons, making it easier to identify destinations at a glance • Composer image attachments open in lightbox – Images attached in the composer now open in full-size preview instead of the document rail • Inline image download button anchoring – Image download buttons are now positioned directly on images for cleaner, more intuitive interaction • Tab shortcuts follow focused zone – Command-W and Command-T now target the active chat tab, making tab management more intuitive with multiple conversations • Fixed thin scrollbars on macOS – Desktop scrollbars on macOS are now consistently thin and visually appealing across all platforms • Chromeless markdown code blocks – Code blocks now appear as simple tinted areas with syntax highlighting, blending naturally with conversations • Camera capture support – Desktop renderer can now access your camera for features requiring video input with proper permission prompts • WebSocket token validation before ready – Desktop checks WebSocket token validity before declaring backend ready, ensuring secure connections from the start • Right-click parity on cron and profile rows – You can now right-click on cron, webhooks, and profile rows to edit or delete, matching existing menu button functionality • Database permission recovery at startup – Hermes now auto-fixes read-only database files at startup or provides clear recovery instructions • Tab strip keeps add button visible – When many tabs are open, the add tab button always stays in view without manual scrolling • Message edit send button works – The send button in the message editor now correctly sends edited messages on desktop • Tips wait before appearing – Tooltip delays are now honored, preventing them from flashing and cluttering the screen as you move your cursor • Keep tab and add button in view – Tab scrolling now ensures the active tab and add button stay visible when you have many tabs • Keep pinned sessions out of unpinned lists – Pinned conversations no longer appear twice in the sidebar, keeping it clean and organized

English
19
16
389
38.3K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
🚀 Ready to play hard and have a blast for less than the cost of oxygen? All eyes are on @deepseek_ai right now with the release of the new V4-Flash model 🔥 Time to jump in and see what it does different! #AIPlay #DeepSeek #TechFun
Sarcastic Badger tweet media
English
0
0
0
52
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@StefanMaier @NVIDIAAI Just read the benchmarks (taken with a grain of salt) from Deepseek themselves, and whichever way you look at it, for a Flash model at that price, it's really impressive! Such a fan of Deepseek, honestly.
English
0
0
0
46
Cesarus
Cesarus@StefanMaier·
If that‘s the case it‘s pretty wild. I immediately could stop my Inkling-small installation. My two @NVIDIAAI DGX Spark can‘t wait to power up the new version of this local llm (when available? 🥹)
Tech2Wild@Tech2Wild

Deep Seek v4 Flash Is BEATING PRO AND GLM 5.2. Models that run on 4+ Sparks. Why use ALL 4 Sparks to Run GLM when you can run 2 Instances of DS4 or better yet run one instance and have 2 Sparks to go do other things with... This is Wild !

English
2
0
9
1.1K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@deepseek_ai Some really impressive scores from my favourite daily driver, BUT, Genuinely curious how V4 Flash handles ambiguous agentic tasks where the "correct" behaviour is contested. Benchmarks measure competence on settled questions. Real agents live in the messy bits. Keen to try it!
English
0
0
1
3.6K
DeepSeek
DeepSeek@deepseek_ai·
🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! 🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇 🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex! Check out the configuration details in our official API docs: api-docs.deepseek.com/quick_start/ag…
DeepSeek tweet media
English
1.4K
3K
24.7K
6.5M
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@gitlawb For those who are lazy to read: 50B-parameter model, consisting of a 35B base model and four 3.7B-parameter LoRA specialists. It is post-trained from Qwen 3.6 and designed for local deployment.
English
0
0
2
179
GitLawb
GitLawb@gitlawb·
Macaron-V1 Tall available now in OpenGateway UNLIMITED & FREE go build your dreams!
GitLawb tweet media
English
22
26
298
17K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@simplifyinAI I use both Hermes Agent and Cowork to write a lot of technical reports for clients, based on data that it monitors for them, and this looks like a great skill to add to them. Definitely giving it a test run!
English
0
0
0
69
Simplifying AI
Simplifying AI@simplifyinAI·
This free directory catches every sign your AI writing sounds fake. It's called Avoid Slop, a curated list of open-source tools that strip AI patterns out of your writing, code, and design. > Humanizer rewrites your text, then audits its own output a second time > Stop Slop bans throat-clearing openers, filler phrases etc. > Unslop studies a model's own habits and builds a custom "avoid this" file for your specific writing > Impeccable fixes AI-designed interfaces that all look the same, wrong grays, no personality Pick one, or run all four together. 100% Free. Open Source.
Simplifying AI tweet media
English
3
8
67
3.2K
tonbi
tonbi@tonbistudio·
There's a lot of talk about which AI model is really the best, so I decided to do a tournament-style bout featuring extremely difficult, multi-hour tasks to find out. The contenders: Claude Fable 5. GPT-5.6-Sol Grok 4.5 Kimi K3 GLM-5.2 Five models enter, one leaves the champ. Part 1 of this two-part series is live now on my youtube channel: @tonbisaigarage" target="_blank" rel="nofollow noopener">youtube.com/@tonbisaigarage
English
18
7
112
6.1K
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
@GeminiApp @aisearchio I told it 6x in different ways to simply stitch 2 videos together in split screen and keep the original videos in the output, and all 6 times I had completely different videos in split screen. Not a great start
English
0
0
0
1.2K
Google Gemini
Google Gemini@GeminiApp·
Starting today, you can create ten videos *at no cost* in Gemini until 11:59pm PT on August 4th, 2026. Just select "Create video" in the tools menu to create, edit, and remix videos to bring your ideas to life. Share how you’re using Gemini Omni in the replies 👇
English
315
505
3.5K
1.5M
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
⚠️ Can any of my mutual connections please verify if you see this post?
English
0
0
0
7
Sarcastic Badger
Sarcastic Badger@Entropy_Badger·
🚨 Big AI breakthrough just dropped. Anthropic’s unreleased Claude Mythos Preview model just did something wild. It discovered two brand new cryptographic attacks on its own. One weakens HAWK, a NIST post quantum signature candidate. The other sharpens attacks on reduced round AES. This is not just another benchmark flex. Frontier AI is now performing original cryptanalysis at the level of top human experts. 🔓 The security world just got a serious wake up call. 🔗 anthropic.com/research/disco… #AINews #CyberSecurity #Anthropic
English
0
0
0
4