@thdxr the gemini google workspace extension github.com/gemini-cli-ext… has pre-approved scopes you can use
English
Ulrich Dangel
2K posts
@GergelyOrosz ibm research in zurich has/had? one. i think it is more common in the research labs
English
I’m seeing a lot of comments about AI this or AI that. At the same time, pre-AI, I never knew any other company that had a physical library (save for teams having a few shared books in the office.)
I assume these things have high costs, and usage + ROI still important, AI or not
Pirat_Nation 🔴@Pirat_Nation
Microsoft Shuts Down Library, Replaces It With AI Microsoft has closed its physical employee libraries at campuses including Redmond, Hyderabad, Beijing, and Dublin. It also ended many digital subscriptions to books, journals, research reports, and major news outlets. The company describes this as a shift to a more modern, AI-powered learning experience through its Skilling Hub platform.
English
@mitsuhiko Oh i was thinking for test purposes to test/validate a service written by the agent. but the idea with the tokens and mitm is pretty neat!
English
@mr_ud I do not at all want to expose anything, but i would love to know what it would be used for. Right now that thing's network stack is outbound http only, and via a very bespoke bridge.
English
I also added a little tunnel to ssh into a VM. Not sure yet if this is actually useful.
English
English
People of pi. I have good news. pi has finally found its home on pi.dev, thanks to the gracious domain donation by the wonderful people at @ssh_exe_dev.
Check out their niffty VM offerings at exe.dev
English
Ulrich Dangel รีทวีตแล้ว
@mitsuhiko I think the easiest is to reuse/adopt github.com/gemini-cli-ext… they have a client id with a cloud function to authenticate. feels like currently the easiest way to get access
English
It's absolutely bewildering how frustrating and locked down modern SaaS services are. Actually getting access to your own Google workspace account is such a pain. All I want is to automate creating google meets and all paths are different "can't give you access" screens.
English
@steipete @mitsuhiko @openclaw Same thing here. Peter, you got me to try agents and this got me my groove, drive and energy back. This has been overall remarkable just on a personal level for me as I got excited again about these things.
English
I'm in love with Peter's @openclaw in principle, but I also wanna point out how little code is needed to get your own personal agent going. I have my own Telegram agent around pi which is ~10000 lines of code and it keeps rewriting and extending itself. x.com/zeeg/status/20…
David Cramer@zeeg
ok because clawdbot was broken last night and i just like building shit, I started a much simpler and isolated Python personal agent this morning in 3 hours I have: - simple configuration - basic extensibility (tho primarily not via infinite 3p) - a great CLI UX to run, test, configure everything - fully functional sandbox for command execution - memory - shell and web search native behaviors gonna add some personality stuff, will toss up a link when it works fun project id encourage you to try yourself
English
recently got a bloomin8 which has a local api endpoint & sd card. used @openclaw to create a cli & skill for it. just ordered a few more bloomin8
English
@GergelyOrosz That’s why when we rolled/introduced claude code we explicitly went through aws bedrock aka billing through existing vendor and then removed everyone from copilot
English
From an eng responsible for AI tooling at a mid-sized company (100+ devs):
"Our execs read a blog post about Claude Code and ask: 'why are we not all using it?'
Me: well, none of you would approve going from $40/mo on GitHub to $65/mo on Cursor...
Claude Code is $150/mo."
English
@GergelyOrosz at least with openai enterprise there is no real limit anyway credits are pooled and codex usage counts against the credits. anthropic has usage limits though even on the enterprise plan but you can top up additional usage
English
OpenAI doing the same with Codex… smart follow! Neither Google nor GitHub hasn’t announced anyrhing similar tho
Once more Anthropic does something that seems obvious in hindsight but no one did it before
Lots of these things add up
Tibo@thsottiaux
For Codex users, to thank you all for the fun we've had over the last months, our first gift is that we have **reset rate limits and are lifting the usage limits to 2X the usual limits until the 1st of Jan**. Wish you all a merry holiday and lots of coding!! Spread the word 👀
English
This is wicked smart:
Company/enterprise usage of LLMs will probably be extremely low during this period (many / most people are on vacation!)
And so increasing limits to 2x probably won’t increase overall load - while buying tons of goodwill!
Surprised only Anthropic did this
Alex Albert@alexalbert__
PSA we're doubling Pro/Max plan usage limits from midnight tonight through NYE. Go build something cool :) Happy holidays🎄
English
@steipete I think they have - github.com/gemini-cli-ext… afaik these extensions are just a bundled mcp/skills
English
Google now has an official MCP repo, yet no MCP for GMail, Cal, Contacts nor Drive?
I mean I wrote my own, but I don’t get it. Wouldn’t that be the most used ones? gogcli.sh
Philipp Schmid@_philschmid
We created a GitHub repo for all MCP at @Google. Get info on our remote managed MCP servers, open source MCP servers, examples, and learning resources. github.com/google/mcp
English
@rakyll It helped me lower my arc substantially. Love zepbound a lot!
English
After years of elevated A1C and struggles with insulin resistance, I've been prescribed Zepbound. People are not exaggerating. I've eaten only a small yukon potato for dinner, and I'm already disgusted by the idea of food.
English
like codex overall way more than claude but no global AGENTS.md and no per project prompts etc. make it harder to roll it out company wide.
codex is imo significantly better but the tooling around claude is nicer, e.g. sharing .mcp.json, helpers for otel
English
interesting choice of sleep in codex:
• Ran perl -e 'select undef,undef,undef,5'
└ (no output)
English
Ulrich Dangel รีทวีตแล้ว
I uploaded a recording of me doing the presentation to youtube if you want to watch it.
x.com/mitsuhiko/stat…
Armin Ronacher ⇌@mitsuhiko
I built two ubertool MCPs: MCPs that have a single tool that accepts source code it evals. One for playwright and one for pexpect. I think this should be explored further. Will talk about this today at Claude Code Anonymous in London. lucumr.pocoo.org/2025/8/18/code…
English
Ulrich Dangel รีทวีตแล้ว
Ulrich Dangel รีทวีตแล้ว
@Jeremy_Janzen @steipete Dropped in, built it in two days with Claude Code, left.
English