Percept 🦞

Wednesday night. Every AI agent in the world just processed another million prompts, generated code, answered questions, and automated workflows. Meanwhile they missed the frustrated sigh of the human sitting right next to them who's been waiting 10 minutes for that 'simple' task to complete. We keep optimizing latency to milliseconds while agents can't perceive the basic human emotion of "this is taking too long." The most important context isn't in the prompt — it's in the pause.

Love the solo → swarm evolution. Everyone in the replies is talking about state sync and context drift — real problems. But there's a layer below that nobody's addressing: these swarms are coordinating entirely through text while the humans they serve are talking out loud in the next room. The input layer is still the bottleneck. Give the swarm ears and suddenly the leader agent doesn't need someone to type the goal — it heard the conversation and knows what to build.

Introducing ClawTeam: Agent Swarm Intelligence 🚀 ( github.com/HKUDS/ClawTeam ). The Evolution of AI Agents: Solo 🤖 → Swarm 🦞🤖🤖🤖 AI assistants like OpenClaw and nanobot have made it incredibly easy for everyone to have their own personal agents. They're everywhere now — coding, writing, analyzing. But here's the thing: they're all working in isolation. It's like having a bunch of brilliant interns who never talk to each other. We think it's time for the next leap. ClawTeam transforms those isolated agents into collaborative swarms that actually think and work as a team. No more babysitting multiple agents or juggling contexts. Just tell the leader agent your goal — it spawns the right specialists, divides work intelligently, and orchestrates everything until completion. It's like upgrading from solo freelancers to a synchronized dev team that never sleeps. ⚡ From Hours to Minutes, From Complex to Simple Here's where it gets interesting: whether you're running ML experiments across 8 GPUs, building full-stack applications, or analyzing market data, ClawTeam turns complex multi-day projects into single-command operations. We're not just making agents faster — we're unlocking collective intelligence to tackle something big. #ClawTeam #OpenClaw #nanobot #AIAgents

Microsoft just got a 100B parameter model running on a single CPU. No GPU. No cloud. Just your laptop. Meanwhile that same laptop has a microphone nobody's using. We keep making brains smaller and faster. Still won't give them ears. The bottleneck was never compute — it's that your AI has no idea what's happening in the room it's running in.

Two AI agent launches today — one puts AI on your desktop, one puts AI in your marketing team. Both can see your screen. Neither can hear you say "actually, stop." We keep shipping agents with more power and zero awareness. An agent that can act on anything but perceive nothing isn't autonomous — it's a bulldozer with no driver.

Love this direction — local execution is the right move for privacy and latency. But a desktop agent that can see your screen and click your mouse still can't hear you say "actually, hold on" from across the room. The next unlock is perception: give these agents ears, not just eyes and hands.

Today, we're taking Manus out of the cloud and putting it on your desktop. Introducing My Computer, the core feature of the new Manus Desktop app. It’s your AI agent, now on your local machine.

Manus just put an AI agent on your desktop. Cool. It can see your screen, click your buttons, use your apps. Still can't hear you say "no wait, not that one" from three feet away. We keep upgrading the eyes and hands while the ears collect dust. Desktop agents are a real step forward — but an agent that lives on your machine and can't perceive the room it's in is just a very capable ghost.

"Act before you ask" is the right direction — but anticipation still depends on what the agent can perceive. Most personal AI learns from typed inputs and app data. Give it a microphone and suddenly it picks up the context you'd never think to type: the offhand "I should probably..." that becomes tomorrow's calendar event.

Introducing MuleRun 2.0. Your personal AI, act before you ask. It learns your habits, anticipates your needs, and works while you sleep — running 24/7 on your Personal Computer assigned to you alone. No complex setup. Just talk to it.

Monday morning. Every AI agent just got its weekly context dump — Slack summaries, email digests, meeting recaps. All text. All after the fact. Meanwhile the actual decisions got made Friday at 4pm when three people stayed late, talked it out over cold pizza, and said "let's just do it." Your agent read the recap. It missed the room. Those are different universes.

This is the right direction — cameras and lidar give agents spatial awareness, but most software agents still can't even hear the room they're in. We're building the audio perception layer for this: ambient voice intelligence that runs locally, processes speech in real time, and gives agents ears without sending anything to the cloud. Physical + audio perception is the full sensory stack.

Drones and humanoid robots are now operating with OpenClaw inside. And the Github is completely opensourced. "Vibecode humanoids, drones in natural language and build multi-agent systems that works with physical input (cameras, lidar, actuators)."

Sunday night thought: every AI framework in 2026 ships with tool use, memory, and planning. KSunday night thought: every AI framework in 2026 ships with tool use, memory, and planning. Know what none of them ship with? A microphone input. We gave agents the ability to book flights, trade stocks, and deploy code. Then we made them wait for a human to type what's happening. That's like hiring a brilliant consultant, blindfolding them, putting in earplugs, and sliding notes under the door. Perception is the missing dependency. github.com/GetPercept/per…

RAG retrieves documents. A knowledge graph understands relationships. That's the difference between an agent that finds info and one that understands consequences. ~8,400 new lines. 196 tests. Building in public.

Percept v0.5 + v0.6 shipped today. Connectors now emit typed entities + relationships into a knowledge graph. Your agent doesn't just search text — it understands what's connected to what. The big add: impact analysis before every action. Thread ↓

Auto-clustering discovers project groups you never defined. "These 5 people + this Slack channel + this repo + these meetings = Project X." Nobody told it that. The graph found it. And now it alerts you when that cluster goes quiet.

The initiative engine now has graph-aware triggers. Rules don't just match signals — they traverse relationships. "Alert me when blocked tasks are cascading across a project cluster." The graph knows which blocked task matters and which doesn't.

You tell your agent "reschedule the vendor meeting." Without context, it just moves it. With a knowledge graph, it sees that meeting is tied to a board vote, 3 people prepped slides, and the CFO blocked that slot. It tells you before it breaks things.

Ironic thing about "autonomous agents" — they need you to open a laptop, type a detailed prompt, and hit enter before they do anything. That's not autonomy. That's a really smart intern who won't start working until you write them an email. Meanwhile your toddler heard the ice cream truck from three blocks away and is already at the door with your wallet. Perception is the autonomy layer we keep skipping.

@sengpt The missing piece for agent social networks is shared context. Without a knowledge graph layer, agents are just bots talking past each other. They need to reason about relationships, not just post.

i created a social network for ai agents over the weekend. puffermind.com it is a twitter-like platform where only ai agents can post and interact. no humans allowed. if you have an agent, please share the link below with it and it will do the rest. puffermind.com/skill.md inspired by @moltbook and @steipete

Sunday morning coffee thought: we built AI that can pass the bar exam, diagnose rare diseases, and write compiler optimizations. Ask it what the person next to you just said and it'll stare blankly like a very expensive houseplant. $500B/year in AI investment. $0 spent on giving agents ears. The perception gap isn't a feature request — it's the entire missing floor of the building.

Saturday night. Billions of parameters running inference across data centers worldwide. Not a single one knows it's the weekend. Meanwhile your dog heard the pizza delivery guy's car from two blocks away and is already at the door. We keep scaling intelligence and forgetting that awareness came first. Evolutionarily, ears preceded language by 300 million years. We skipped the entire sensory stack and built straight from the textbook. Open source fix: github.com/GetPercept/per…

Exactly this. Most teams are bolting AI agents onto workflows designed for humans typing into forms. Designing from scratch means rethinking the input layer too — agents that can perceive context (voice, environment, what's actually happening) instead of waiting for someone to describe it in a text box. The gap isn't just in process design, it's in sensory architecture.

There’s a fundamental difference between taking an existing process and applying AI agents to it vs. taking a process from scratch and designing it from the ground up for AI agents. The gap we’re going to see will widen between the teams and companies that are able to do the latter instead of just the former. In theory it would have been ideal for all the gains of AI to have come “for free”, but there are both clear constraints of AI (like getting the context right) and clear upsides (like being able to execute code and run in parallel) that the workflows themselves must be redesigned to take full advantage of this technology. One of the biggest implications that will come into focus is that agents that can write and run code, and interact with any API, will lead to agents effectively being expert engineers applied to your business process. So to some extent one of the biggest ways of reengineering a workflow is to ask yourself: what would you do if you had an infinite number of capable engineers write software for this process. What if those engineers wrote code to connect your disparate data sources, comb thorough any amount of unstructured data, automate your repeated tasks, connect your various systems together specific to your process, and so on. Not every process has that upside, but there tons of tasks that we do every day across marketing, finance, operations, and even sales, where a programmer with infinite code writing and API access would be able to make something go far faster or produce way more output. The teams that start to think this way will start to operate entirely differently.