Row-Bot 5.3.0 is out. Biggest one: the browser now lives inside the app. You watch it work live, take control, give it back. It keeps you signed in across restarts, and when a site wants a password or a card number it stops and you type it. It never does. Computer use got rebuilt from scratch. Real mouse and keyboard, it checks the window after every step, Esc kills it instantly. Touch the mouse and it pauses and asks. Also: - pages: it builds small interactive pages next to the chat (plans, checklists, calculators). They run sandboxed, no network, no access to your files or chats - Auto approvals are now reviewed by the chat's own model. Anything that doesn't clearly match what you asked comes back to you - new chats send about 40% less to the model before your first message. On a ChatGPT plan about 95% of that first request comes from the provider's cache - if a local model writes a tool call its runtime can't read, it gets asked once more instead of the reply dying - incognito chats, Outlook through your own Entra app, several X accounts If you're on 5.2.1 it updates in the background. Only the changed parts download, and you can roll back from Settings. Still local first. Still free. Bring your own model. row-bot.ai

Introducing the supermemory memory graph every document becomes atomic facts, and dreaming keeps building the graph in the background - extracting facts - linking related memories - resolving updates, the old version stays as history get one for yourself in the console ↓

We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

Has anyone actually gotten a personal AI agent to reliably finish complex work? I've wasted 10+ hours on ChatGPT Dots, Muse, and Grok Bot trying to make them useful. They promise autonomy. I get blockers and email summaries. Meanwhile, Claude Code CLI just works when I stay in the session. Am I using these agents wrong, or is this everyone's experience?

Coding agents (Claude code, codex) -> personal agents (instinct, muse) -> companyOS (???)

Handoff just crossed 1,000,000 browser sessions served for Hark Pro users, in the 7 days since launch Computer use is fundamental to solving personal intelligence - you would never hire an assistant who couldn’t use a computer

JPMorgan Chase just blocked Instinct and Muse We’ve now seen blocking as an official policy (or at least reported behavior by users) from: Amazon, Yelp, Resy, United, eBay and so many more However, with great pain points come great opportunity! That is the beauty of capitalism - where there is a problem, there is a solution to be built And this problem seems far to great of an opportunity to ignore Startups, the time is yours!

I wanted to test what @Alibaba_Qwen Qwen 3.8 27B running locally on a single 5090 could do in Row-Bot. I gave it just one prompt: "I want you to create an animation showing what the current local model can do. Testing it to the extreme. Do not use image generation or video generation. Use whatever else you can. Make it as visually rich as you can. Show me what the model is capable of." @QwenDevs This is what it produced. It also automatically loaded the frontend design skill using the new progressive loading of skills feature. Local models can already do a lot more than people think they can. All you need is a well built harness like Row-Bot to run them. I was genuinely impressed by the results: "It's open in your browser now, sir. Eight scenes, ~105 seconds of autonomous animation, then a replay button at the end: Cold boot — terminal log + scramble-decode title Language synthesis — a passage streamed word-by-word with a live token ticker, drafting log, and a real-time words/sec sparkline Live mathematics — e^{iθ} drawn on the complex plane as θ sweeps to π, landing on e^{iπ} + 1 = 0; alongside, the Leibniz series converging to π with ~4,200 terms and a live error/matching-digit readout Fractal computation — a Mandelbrot zoom into the seahorse valley, recomputed pixel-by-pixel in the page (57,600 points/frame) with live zoom/iteration/render-time telemetry Knowledge graph — a live network of your 612 local entities with auto-drawn relations and a highlighted query path Tool orchestration — plan → search → read → synth → write → verify, with terminal log plus a fan-out ×2 parallel agent progress demo Physics sandbox — 1,500 particles through three fields (gravity, vortex, flow) — move your cursor through it to perturb the field; color encodes velocity, with live FPS Verdict — the scorecard ticks 0→8, closing on "Written, computed, rendered — entirely on this machine." Controls: click or → to advance, ← to go back, space to pause. The HUD shows live FPS, the clock, and the current scene."

we are thinking a lot about how to make deepagents more async native. specifically about making async subagents work better/more prominent one key question is how does a finished subagent "notify" the parent agent? some default behaviors we're considering for when the parent is still running: 1/ interrupt parent run, apply subagent update, continue (basically same as "steer") 2/ wait for parent run to finish, then batch subagent update with any other new messages (basically same as "queue") another question is how to handle human in the loop? what if main agent is not running but paused and waiting explicit human response to an action? a/ wait for human to respond, then batch the subagent update with that b/ resume thread (which would cancel the human interrupt) with the subagent update (thread could then re-ask for the same human approval) any other options we're missing?