And Pi out of the box sends less than 1.2k tokens ;) systima.ai/blog/claude-co…

@pidotdev and pi lets you finely see and control what appears in context github.com/tmustier/pine-… or vibe your own extension!

@pidotdev ily pi

@pidotdev Is grok login possible in pi?

@pidotdev why no plan mode in pi like the opencode

@pidotdev I really feel OpenCode is terrible harness... would choose pi any day over it. If you try minimax in pi/claude vs OC, the model feels extremely dumb in OC compared to others

@pidotdev Any idea if we can run Pi in browser? Trying to embed it in a chrome extension

@pidotdev Pi needs to give opencode equivalent packages from official sources. 3rd party plugins are just too risky.

@pidotdev You would be shocked at how well this plugin performs but it will fuck up compaction if you are using Claude because it automatically compacts thinking tokens. github.com/entrepeneur4ly…

@pidotdev i wasnt satisfied with that either , so i implemented this github.com/corporatepiyus…

@pidotdev @pidotdev

@pidotdev that is just ... the way of Pi, not only less token consumption at starting time, but also in the whole session. no clear metrics, but i can feel that totally after 1 day working fully with Pi and another day fully with Codex.

@pidotdev tldr; don't let the ice cream seller buy ice creams on your behalf

@pidotdev Pi is the real deal 📟

@pidotdev That's the clean minimalism we like to see – no bloat, just straight performance.

@pidotdev 👑

@pidotdev As a designer, modular is always better. Minimalism always wins long term.

@pidotdev Token saving should be a strategy and not an harness tool. This is why I built HarnessTrim. github.com/giuliastro/Har…

@pidotdev @AkitaOnRails já viu esse artigo?

@pidotdev Tried Pi on whim not thinking much of it but it turned opus in Claude code into a super focused code machine.. hats off to the team it’s my go to harness now

@pidotdev That's a vanity metric. why? * If you are calling multiple prompts, that initial tokens will get cached * 1.2k is nothing when doing some actual work like reading files & etc. So, having better tools is way better than the initial token count.

@pidotdev I found cursor debug mode very strong. Do you have same thing in extensions?

@pidotdev @thdxr is this true?

@pidotdev this ext use a Probe Mechanism to know exactly the real current context give it a try github.com/RunMintOn/pi-c…

@pidotdev My Pi can send 800 :D

@pidotdev what do we think about this ? Empryo vs PI empryo.com/benchmarks

@pidotdev Pi is really very light

@pidotdev pi still has the biggest pipi!

@pidotdev pi should copy claude code prompt for coding area ，for better coding engi neer

@pidotdev 1.2k tokens is lean, but pi still pays when tool calls loop

@pidotdev I use open code with oh my open agent and the context sometimes reach 50-100K just from the first message, crazy shit

@pidotdev There's a missing login method for grok subscriptions (inclWe still don't have a login method for grok subscriptions (including X Login) set up yet 🤔

@pidotdev If u need code review like cc /code-review ,try👇 x.com/_naiVe_2/statu…