I turned Comet into a no code agent building app. Agents, sub-agents, memory, the works. It uses a setup like Claude Code, and it's insane. No MCPs, no complex integrations. Consumer friendly. How + templates below Add this feature plz cc @AravSrinivas @tonygwu

How does it work? With notes. It dawned on me that browsers AND notes are the IDE for non-coders. Which made me think - what if notes worked like Claude Code inside Comet? So I built a custom notes app, focused on agent use (Comet should just build this in).

Running it this way makes it SUPER easy to setup and tune because it's just text. No visual builders, nodes, or other nonsense. But it still supports agents, lets agents run sub-agents, run reviews, can write memory, or even new notes.

I've got two agent teams live today with it. LinkedIn Agent Team - find opportunities based on interactions. Website Optimization team - weekly detailed analysis, summary and recommendations for a/b tests on the site. Its memory across sessions makes it more powerful. More coming.

Example of the LinkedIn agent team in it: - Playbook (orchestrator): routes tasks, sets order, enforces quality gates - Research Agent: looks up posts and comments from the people I ask it to - Agent Memory: saves what the research agent found - Historian Agent: prevents repeats; checks topic/format collisions from memory - Ghostwriter Draft Agent: drafts messages, writes a fresh "review" note with tags for me to approve - Ghostwriter Apply Agent: executes my approved note, sending messages for me

I honestly wasn't sure if I could get it to work, but after about ~14 hours, it does! Results so far are awesome. ~4 hrs/week saved on ops so far Tees up what I want for review. I use shortcuts once review is done. 1 prompt now creates all the agents (attached)

@PerplexityComet team, steal this 🙏 Build notes into Comet and this could be even more powerful. Adding this in would be like having Claude Code agents for the consumer vs. dev.