Mitch Fultz

If you use pi, try pi-agent-browser-native. It makes agent-browser a native pi tool, so agents actually use browser automation instead of awkward bash glue. In my runs it has materially improved tool uptake, speed, and token-efficient browser work. github.com/fitchmultz/pi-…

@thsottiaux I imagine it's impossible to stop "working." If you're like me it doesn't even feel like work. It's addicting and fun.

Ship => Feedback => Ship => Feedback => Ship => Oops something not quite right => Reset => Yay => Ship => … The never ending cycle of Codex improvements

@thsottiaux Just ship every day.

I don't know why we ship on Saturdays now, but here are a bunch of nice improvements for Codex. Didn't feel right to hold this back until Tuesday when it all makes it a lot more delightful to use.

@RhysSullivan Have you asked Codex to reverse engineer codex?

i really wish the codex app was open source, so many small tweaks i want to make to it to better suit my workflow genuinely really well made piece of software it feels so smooth, i was unfamiliar with openai's game

@Howaboua Did you modify this intentionally? What does pi use by default?

I've just realised I've been using Spark for compaction in Pi. This might not be the best idea. Yes, it uses up a separate quota and is super fast, but I am worried that it might be too dumb. Thoughts? Should I switch to mini?

Just shipped pi-agent-browser-native v0.2.28 🚀 Pi agents now get better real-browser workflows with: - no-op scroll diagnostics for dense dashboards - combobox recovery hints when clicks only focus the field - early ffmpeg warnings for recordings - leaner runtime guidance with on-demand docs - stronger secret redaction Built for agents driving real browser sessions with native tools. pi install npm:pi-agent-browser-native github.com/fitchmultz/pi-… @vercel @vercel_dev

Just shipped pi-cursor-sdk v0.1.10 🚀 Cursor SDK models now feel even more native in pi: - Cursor edit/write activity shows as clean tool cards - no double-running file mutations - safer read transcript fallbacks - better prompt budgeting - live /cursor-refresh-models - snapshot refresh tooling for maintainers pi install npm:pi-cursor-sdk github.com/fitchmultz/pi-…

@jwswj Now I can remove my github token from the cloud env. Great new feature.

Multi-repo environments!

Hey @thsottiaux, could Codex add a menu option to change a project’s working directory? Codex created a new project under Documents, but I wanted it in Projects. After moving the folder manually, my sessions disappeared.

@_gannon_ I couldn't tell you. Not something I've tried. I will add it to my roadmap to investigate.

@fitchmultz Very cool. Starred for later. Does it speak electron? I have a skill for working with electron apps but it remains the most brittle aspect of agent-browser.

If you use pi, try pi-agent-browser-native. It makes agent-browser a native pi tool, so agents actually use browser automation instead of awkward bash glue. In my runs it has materially improved tool uptake, speed, and token-efficient browser work. github.com/fitchmultz/pi-…

@syabro Have you seen the amount of escape errors models make when using bash tools? The alleviates a lot of that pain

@fitchmultz "awkward bash glue" lol "stable easy-to-trace CLI api" you mean?

@micLivs The tool version is far more reliable than using the bash version directly. The constant malformed tool calls with the bash version is what spawned the idea to make this. It's built to be more efficient and reliable. I am working on some further improvements now.

@fitchmultz Obviously :) but does it stumble a bunch before it gets stuff right or does it one shot stuff successfully

No skill file needed. This exposes agent_browser as a tool like read, write, bash and models will automatically invoke it.

@TheFelikZ No skill files needed. This shows up as a native tool like read, write, bash, etc.

@fitchmultz Whats the benefits of using this compare to playwright-cli + skill? Yes it wraps tools to “native” tools, however the agent still have to draft tool call which is similar to drafting a CLI call

@micLivs No skill use at all. Since pi sees agent-browser as a native tool it automatically invokes it like it does the read, write, and other pi native tools.

@fitchmultz Browser use is so messy, did you manager to avoid skills?

@bennett2b Automations are incredibly useful. CI gates and actions are stuck with deterministic outcomes while Automations unleash a whole new world of possibilities.

Spent time with a customer and an FDE on our team yesterday building Cursor Automations built on Cloud Agents. They estimated that one of the Automations we built will return 40 hours per week to their team. Insane. And we built 20 other Automations in 2 hours.

@danyel117_ @lucasmeijer Reduce token consumption further with github.com/fitchmultz/pi-…

@lucasmeijer Agent browser by Vercel has been doing the job for me lately. Very good and not so token-consuming. github.com/vercel-labs/ag…

i've been using pi + playwright + chrome, to close the loop for my agents on webapps. Is this still where it's at? are people using that chrome extension thing codex uses for computer use?

@bedesqui @lucasmeijer Add agent-browser a pi native tool. github.com/fitchmultz/pi-…

@lucasmeijer use pi + agent-browser, its so good

@meln1k @lucasmeijer Make agent-browser a pi native tool. github.com/fitchmultz/pi-…

@lucasmeijer I'm find agent-browser from vercel to work very well with pi.

@happy2png @lucasmeijer Take it one step further. github.com/fitchmultz/pi-…

@lucasmeijer have you tried agent-browser from vercel using that atm with pi

@Anton_Kuzmen @lucasmeijer Make it even better. Add agent-browser as a tool. github.com/fitchmultz/pi-…

@lucasmeijer agent-browser + pi is the best browser automation yet. this is literally the only skill I have in pi (you can find it in their repo vercel-labs/agent-browser)