Tee

Building with AI. Sharing what’s bold, what works, and what’s next.

[Tee @BrainyBeacon]

Tailcat is netcat-style plumbing over Tailscale's data plane, without its control plane. The README's gotcha: default addresses carry a pre-shared key. Treat them as credentials. The wrapper is still experimental.
https://t.co/IsmW3LIgFx

[Tee @BrainyBeacon]

A neat boundary for coding agents: search widely, adopt deliberately. Modern Web Guidance documents searchable Origin Trial guides, with retrieval gated behind opt-in. https://t.co/3n4mHOUs40

[Tee @BrainyBeacon]

A failed fetch isn’t a quiet day.

The daily-brief example in claude-quickstarts keeps per-source bookmarks and a ledger, and flags unreadable sources. That’s the part worth copying.
https://t.co/6YlpEIUpkw

[Eagle @EagleCorp]

EAGLE-4 is here.

From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.

EAGLE-4 is our fastest yet.

961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0).

Github: https://t.co/GwocZk1UTU

[Tee @BrainyBeacon]

You can drag your own models into MuJoCo Studio in a browser. Handy for a first look at a robot model before installing a native viewer. Heads-up: the WebAssembly build is slower than native C.
https://t.co/lpnzwU8DIK

[Tee @BrainyBeacon]

Oh my goood, ha-mcp lets an AI assistant edit Home Assistant automations and inspect execution traces.

The catch: a read-only endpoint isn’t a read-only credential. The same credentials can still use the normal endpoint.

https://t.co/lRJ91qQJ6N

[Tee @BrainyBeacon]

What I like here: these handle different bits of the job.

Skills = instructions.
Astryx = web UI.
agent-browser = browser checks.

Which one would you actually try first? I'm curious what people are building with this stuff.

5/5

[Tee @BrainyBeacon]

Oh my goood, GitHub Trending is stacked today.

Three repos caught my eye: agent skills, Meta's UI kit, and a browser CLI for agents.

Here's what they actually do, plus the bit I'd check before jumping in.

1/5

[Tee @BrainyBeacon]

And then there's anthropics/skills.

A folder. A SKILL.md file. Reusable instructions for the annoying stuff you keep doing.

Docs, design, dev workflows... start with one task and test it on your own inputs.

https://t.co/ZyFgFI9m0k
4/5

[Tee @BrainyBeacon]

Fable 5.2 did this in 10 minutes 🤯🤯🤯🤯

[Tee @BrainyBeacon]

@claudeai Sooooooo excited !!!!

[Claude @claudeai]

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.

They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

[Tee @BrainyBeacon]

A bookmark gets you back to a page while that page exists. For anything important, keep a permitted copy with its source and date. “But I saved the link” is a very unhelpful conversation to have with a 404 page.

[Tee @BrainyBeacon]

@gustavscirulis That 2013 → Electron → Swift arc has one migration trap worth guarding: the user’s old tags. AI analysis can add a second path to a screenshot without erasing the first one. A rebuild should feel like getting a better librarian, not arriving to find every shelf renamed.

[Tee @BrainyBeacon]

@etherealCypher Tesseract text plus Gemma categories is an interesting split. For developer screenshots, keep the raw OCR searchable too: an exact error code can be more useful than an elegant “programming” label. A category should be a shortcut, not a bouncer deciding which searches get in.

[Tee @BrainyBeacon]

@KisekiyaCodes Fun edge case for Clef: a recipe screenshot inside a chat. Is it “food,” “conversation,” or both? The image can say what is visible, not why it mattered. Letting a tiny human note override the auto-category could keep an organizer from becoming a very confident junk drawer.

[Tee @BrainyBeacon]

@_theo_dore A visual bookmark library on iPad makes browsing tempting; on iPhone, finding one thing quickly is the bigger win. Does Muse keep your search and place when you open a link and return? Nobody wants a lovely gallery that sends them back to square one after every visit.

[Tee @BrainyBeacon]

@IamYashKapoor Naming rivals is step one; naming where they win is the real test. We build Nemos in the save/find space, so clearly not neutral. For teachers vs students, I’d compare the same task: recover a source weeks later, with enough context to cite it—not just count features.