News for hackers. Stuff that matters.

picklepixel

Gemini is really special compared to the other models
Re: on Wednesday February 18, 2026 @10:49PM (#47069676)
I'm actively job hunting and it was eating all my time, so I decided to automate the hell out of it.

ApplyPilot is a 6-stage pipeline from discovering jobs to applying to them. I kept it running for 2 days and hit 1,000 applicationsmostly remote, some local, all in tech.

Some things that genuinely blew my mind so Haiku is the brains of the operation. At some point a LinkedIn session expired mid application but Haiku reset the password and kept going. Another job had no application form at all, just a contact email buried in the description but then it sent the company an email directly and attached my resume. One application was entirely in French and yet it completed the whole thing in French. It reasons through whatever it encounters.

Currently in interviews because of this. Open source: https://github.com/Pickle-Pixel/ApplyPilot

How do you manage context window?

I have been experimenting with context window on Claude Code, and I have stopped using /compact and instead i document the journey well including writing in claude.md.but sometimes there is just some context that i'd like to keep with my next sessions but i do not want to burn tokens just to keep it, so i made an MCP server that does session recap through different models, and spits back a...

( Read More... | no comments | 2 points | on HN )
Re:Speed up responses with fast mode on Sunday February 08, 2026 @12:13AM (#46931501)
or you can proxy Gemini through Claude Code
Re:Stop Paying for API Tokens on Friday February 06, 2026 @09:46AM (#46913456)
TLDR; use your subscription as an API key, powerup your agent.
Re:Stop Paying for API Tokens on Friday February 06, 2026 @09:46AM (#46913447)
I built this because I pay for multiple subscriptions and I want to use them all in the same place at the same time. HydraMCP an MCP server that connects different LLM providers such as ChatGPT, Gemini, Claude and any local model(if you got the hardware) as a tool to be accessed by Claude Code, but here is the catch, you can use your subscription based account for any of these providers instead of an API key and therefore save yourself lots of money. basically use what you pay for on your own way.

five tools exist: you can ask a model, compare models in parallel with side by side metrics, poll models for consensus with an LLM judge to evaluate, synthesize responses from multiple models into a combined answer, list all accessible models by your providers.

happy quota burning!

Re:I Gave Claude Code Infinity Gauntlet of LLMs on Thursday February 05, 2026 @11:12PM (#46909028)
better TLDR; use subscription based LLM providers as API keys.....
Re:I Gave Claude Code Infinity Gauntlet of LLMs on Thursday February 05, 2026 @10:52PM (#46908882)
TLDR; I gave Claude Code super powers, and so can you!
Re:I Gave Claude Code Infinity Gauntlet of LLMs on Thursday February 05, 2026 @10:52PM (#46908880)
I built this because I pay for multiple subscriptions and I want to use them all in the same place at the same time.

HydraMCP an MCP server that connects different LLM providers such as ChatGPT, Gemini, Claude and any local model(if you got the hardware) as a tool to be accessed by Claude Code, but here is the catch, you can use your subscription based account for any of these providers instead of an API key and therefore save yourself lots of money. basically use what you pay for on your own way.

five tools exist: you can ask a model, compare models in parallel with side by side metrics, poll models for consensus with an LLM judge to evaluate, synthesize responses from multiple models into a combined answer, list all accessible models by your providers.

happy quota burning!

Re:I reverse-engineered Netflix's 4K restrictions on Wednesday January 28, 2026 @07:16PM (#46803776)
TLDR; I made an extension to force 4K on Netflix.
Re:I reverse-engineered Netflix's 4K restrictions on Wednesday January 28, 2026 @06:46PM (#46803459)
I pay for Netflix Premium but was stuck at 1080p. Turns out Netflix layers multiple capability checks before serving 4K: user agent, screen resolution, Media Capabilities API, codec support, DRM robustness negotiation, and their Cadmium player's internal bitrate caps.

Built an extension that spoofs all of these. The interesting discovery: you have to intercept every layer. Miss one and you're back to 1080p.

Here's the catch though. Even with all the JavaScript spoofs working, Chrome still won't get 4K. Netflix requires Widevine L1 (hardware DRM), and Chrome only has L3 (software). The browser literally can't negotiate the security level Netflix wants. Edge on Windows has L1, so the extension actually delivers 4K there.

So what's the point on Chrome? Honestly, not much for 4K specifically. But the reverse-engineering was the interesting part. Understanding how Netflix fingerprints devices and decides what quality to serve. The codebase documents all the APIs they check.

On Edge: works reliably, getting 3840x2160 at 15000+ kbps. On Chrome: spoofs work, DRM negotiation fails, stuck at 1080p.

The repo has detailed documentation on what each spoof does and why. Happy to discuss the technical approach or answer questions.