I tried it with Claude Code for a while but lack of WebSearch tool became a dealbreaker for me. Does anyone know of they will provide support for it?

You can integarte a search mcp server. I use it this way and it works flawlessly well.

I don't know why I didn't of this before. Thanks for the suggestion.

ur wc :)

They could have saved themselves 3MB by converting it to AVIF.

For most static UI surfaces, I probably wouldn't use it, but I can see a use case in this for testing generative UI workloads.

Sure, server-side or client-side generated UIs tend to have a lot more interesting complexity to test. But I do want to bring up that with the specification language being TypeScript, you can validate some basic properties even for statically generated sites. I wrote a spell checker for Bombadil that uses https://github.com/wooorm/nspell and a custom dictionary and found a bunch of old typos on my own blog.

I'm curious what a "stripped down version" of Github can offer in terms of functionality that Github does not? Is it not simpler to have the agents register as Github repos since the infrastructure is already in place?

The key idea is not running constantly, but being always on, and being able to react to external events, not just your chat input. So you can set a claw up to do something every time you get a call.

Is AssemblyAI or Deepgram compatible with OpenAI Realtime API, esp. around voice activity detection and turn taking? How do you implement those?

I am not using speech to speech APIs like OpenAI, but it would be easy to swap the STT + LLM + TTS to using Realtime (or Gemini Live API for that matter).

OpenAI realtime voices are really bad though, so you can also configure your session to accept AUDIO and output TEXT, and then use any TTS provider (like ElevenLabs or InWord.ai, my favorite for cost) so generate the audio.

This reads like a classic case of bureaucratic mission creep. When the budget depends on showing numbers, and the definition of a "threat" is loose enough, the agency naturally evolves into a political tool. The lack of uniform identification and masks just removes the final psychological barrier of accountability for the individual agents.

I'm curious how the unit economics actually play out here compared to traditional search. With Google, the compute cost to serve a query is negligible, so even low-CPM ads are profitable.

With an LLM, the inference cost per query is orders of magnitude higher. Unless thy have a way to command significantly higher CPMs -- perhaps by arguing intent signal is stringer in a conversation than a keyword search -- it feels like a difficult margin to sustain.

This man is a legend. He's been working on making a better marble machine for more than a decade. The only payoff is the satisfaction of listening to the synchronized syncopation of falling marbles.

Here is a more recent video: https://www.youtube.com/watch?v=k8NXF2rtaEg

Happy thanksgiving all. In an era where algorithms on other platforms seem optimized for outrage and engagement bait, I'm grateful for HN's optimization for curiosity. It's one of the few places left where I can open a thread on a topic I disagree with and actually expect to have my mind changes -- or at least understand the opposing view better -- by the top comment.