Vatsal Mishra

🛡️I built FakeOut - An app that detects AI-generated images/videos and fact-checks viral claims in seconds. 12K+ downloads and counting. In a world flooded with deepfakes, FakeOut helps you spot the real from the fake. Try it now: fakeout.io 1/4

TTS without a GPU, without an API key, without a monthly bill. 25MB model, runs on CPU, 8 voices, open source. KittenTTS. github.com/KittenML/Kitte…

@shiri_shh Same as Codexbar by @steipete or is it different? github.com/steipete/codex…

someone made a macOS menu bar app that shows your Claude usage.

@karpathy Love Dobby the house elf claw haha!! I have Dr. Cloidberg

Thank you Jensen and NVIDIA! She’s a real beauty! I was told I’d be getting a secret gift, with a hint that it requires 20 amps. (So I knew it had to be good). She’ll make for a beautiful, spacious home for my Dobby the House Elf claw, among lots of other tinkering, thank you!!

@cyrilgupta The tweet is about Anthropic, but the video is Google’s Personal Intelligence😅

Big move by ANTHROPIC🔥 Your phone is now a remote for an AI worker on your computer.

@dansemperepico If you’re using Opus /Sonnet 4.6 then could be an Anthropic thing..it’s been slow for me too since last night..and Opus was down earlier today as well

My OpenClaw is so unbelievably slow now. I mainly use it for information capture, quick voice note yapping to turn into written posts, and food/workout tracking. I just gave it a very short text to edit and it took 4 minutes to reply. Anybody else experiencing this?

@MatthewBerman How does it compare to Opus 4.6?

getting there...

@zarazhangrui Agreed! That's actually how I've been learning about agents and orchestration, by just talking to my openclaw agent every day about how it works lol. Such a great way to learn!

If you wanna level up in AI, spend an hour every day on tinkering and playing with the models. Build something & have fun with it. Test the model’s limits. Don’t be utilitarian. I guarantee that if you do this consistently for a month, at the end you’ll look back and realize this was the most productive way to spend your time. Not reading about it. Not taking a course on it. PLAYING with it. To quote @steipete: it’s hard compete with somebody who’s just there to have fun

WAIT!!! ALL THIS WAS JUST LAST WEEK IN TECH???

@Devinbuild I've been thinking of trying something like this! What models are you using and what are the API costs like?

While I’m traveling my OpenClaw is working 24/7 🦞 I vibe coded an app to help prediction market traders I’m communicating via slack with Nova my chief of staff who is delegating work to the rest of the team The goal: Scale the app to $10k MRR

@steipete @kumareth @steipete after getting bombarded with AI slop but with typos lol

@kumareth I get sooooo much AI stuff my way so I stopped reading these. If you want my attention add some typos :)

Hey @steipete, as you said DenchClaw would make a great plugin, I answered elaborately here on why it wouldn’t be possible under the current structure:

@ollama @Scobleizer @steipete Excited to try this out. Although my understanding is we shouldn’t run smaller local models as the main agent because of prompt injection risks, BUT there are still subagents which can be useful to run locally? Anyone have any great use cases for local models as subagents?

Ollama is now an official provider for OpenClaw. openclaw onboard --auth-choice ollama All models from Ollama will work seamlessly with OpenClaw. 🦞 Use it for the tasks you want, all from your chat app. Thank you @steipete for helping and reviewing. 🦞

@steipete that’s fair, thanks for responding ! So there’s an appeal process, that’s good to know haha

@cryptikcell folks email me and I eval, usually it was shitty behavior and AI was right.

My openclaw twitter mention block cron job is working unreasonably well. Turns out AI is really good at detecting spam/reply guy/promo stuff. Runs every 5 min and cleans up my mentions - I actually see useful replies now and Twitter got pleasant again!

I just asked my @openclaw to alert me when we’re in 2x mode for Claude in my time zone so that I can do the heavy tasks during that time. This is so much better than going to some web app to find out the time when it is 2x in my time zone..personal AI assistants make such thins so easy!

@andrewbrown Hahah..I wouldn't be surprised if it's all Claude doing the launches and marketing lol

@cryptikcell You probably can't at this point. Anthropic seems to be shipping things in an odd order. Maybe there are no humans at the helm

Anthropic just announced Claude Certified Architect exam. Aren't you glad I started my Claude certification course last week? I just knew it my bones, that I had to make one, and now I can just align mine to this.

@daniel_dhawan @katedeyneka It looks quite good! Do you prompt it to tell what do you want to convey in the video or is that also part of the service offering?

Tried Reelful, and it made a great video about Rork office life from pictures of our team Proud that Reelful app was created in Rork! Great job @katedeyneka

@frankdilo @openclaw Love the group setup. My wife and I are using separate Openclaw chats haha. It's still a single instance but both of our sessions, memories etc. are isolated.

🦀 I've been using @OpenClaw daily for weeks now. Here's what I actually do with it. 📬 Morning brief Every day at 6am, it sends me weather + a digest of newsletters from my RSS reader (Feedbin). Summarizes Matt Levine's Money Stuff so I get the key points in 2 min. Also includes my calendar events. 📅 Time blocking "Help me time block today. Max 8h work, I have 3 meetings." It checks all my calendars, asks about priorities, proposes a schedule, and creates the events. Took 5 mins this morning. 👥 1:1 prep Before team meetings, I ask it to help me prepare. It reads my notes, identifies patterns in past feedback, suggests framing, and prepares responses for topics they might raise. 🎤 Voice → Notes I record voice memos with ideas. It transcribes, cleans up, and saves to my notes folder (synced via Syncthing to all devices). Quite useful to get rough ideas out to feed into agents when at the desk. 🐦 Typefully drafts "Create a draft about X" — and it posts to my Typefully queue. I review before publishing. Uses our Typefully skill. ⌨️ Launch Codex/ClaudeCode sessions “Start a Codex session to apply the learnings from this blog post to the Typefully backend codebase” — 3 hours later all your tests are refactored! Can even control multiple sessions at once. 🔌 Tools it controls This is what makes it actually useful vs just a chatbot: - Syncthing (file sync across devices) - Google Calendar (via gog CLI) - Feedbin (RSS reader API) - Typefully (social scheduling API) - Whisper (voice transcription) 👫 Group setup I also have a separate OpenClaw workspace shared with my partner in a Telegram group. You can do this with just one OpenClaw instance since it supports multiple “workspaces”. It has its own identity, memory, and skills. We use it for: - Shared grocery lists - Meal planning - Booking restaurants - Trips research - Household coordination

411 people bookmarked my HEARTBEAT.md tweet yesterday. so here's the advanced version - 4 things most people don't set up that make OpenClaw actually useful overnight: 1. heartbeat schedule - mine runs every 30 mins, 8am to 11pm. outside those hours it stays quiet. without a schedule your agent either burns tokens all night or does nothing. 2. memory flush - before every context compaction, the agent writes what it learned to a daily file. without this you lose everything every few hours. 3. cron jobs for scanning - don't rely on heartbeats for everything. separate cron jobs for heavy tasks like content scanning so they run in isolated sessions and don't bloat your main context. 4. fallback alerts - if a cron fails silently you'll never know. heartbeat checks for failed jobs and flags them before they stack up. none of this is complicated. it's just the stuff nobody tells you to set up on day one.

@linuz90 @openclaw What's the advantage of doing this in Telegram vs Discord channels?

If you're using @openclaw on Telegram, you need to set up forum topics. Turns one messy chat into organized topics. Each with its own isolated LLM session. Each topic gets its own LLM session too 👌 Setup is easy: Open BotFather and enable "Threaded Mode" for your bot. Then tell your agent: → Set up forum topics in our chat. Call getForumTopicIconStickers to get available emoji IDs, then createForumTopic for each one with icon_custom_emoji_id. Important: send a message inside each topic right after creating it, otherwise it won't show up in Telegram. Then after you're happy with the topics, tell it to route all cron jobs and heartbeat messages to the most relevant topic. Now when I forward emails to my bot it acts on them in relevant topics. Sim race prep lands in Racing. Morning briefing goes to Life. Completely changed my experience with it.