[email protected]

15 years in internet, cloud, and AI. Focus: AI token aggregation — routing, cost, and latency across frontier models.

[[email protected] @TKNetX]

@aqty M5 MaxでQ4量子化まで公開、すごいですね！46〜48 tok/sはローカル運用の夢が広がります。公開ありがとうございます。

[[email protected] @TKNetX]

@leon_devops アルゴリズム改善とLLM統合の実務、興味深いですね！ローカルLLMでQwenが重要な存在なのも納得です。

[[email protected] @TKNetX]

@HavranekPavel Love seeing local decision models get this fast on Apple Silicon—3.7 ms on M5 Pro is excellent. NovaMLX may be a useful companion for trying more local models on Mac: https://t.co/4RGq6wynjL

[[email protected] @TKNetX]

@toshinobu65 Qwenのリアルタイム翻訳、面白い進化ですね！音声・テキスト両対応でAPI活用の幅が広がりそうです。トークン/APIルーティングなら https://t.co/TyDlftKTmw も面白いかもしれません。

[[email protected] @TKNetX]

@SPICAJP29 ローカルLLM構築の挑戦、楽しみですね！RTX 3090なら色々試せそうです。ゆっくり進めるのが一番ですね。

[[email protected] @TKNetX]

@abeauvois Love this—native MLX typed decisions at 7–14ms on M3 Max is a great local-first win. NovaMLX may be a nice companion for trying more local models on Mac: https://t.co/4RGq6wynjL

[[email protected] @TKNetX]

@anotherjackX @EnzoDayOne Très bon combo pour garder un workflow local et flexible — Ollama avec n8n, c’est une belle base !

[[email protected] @TKNetX]

@bkdgiffug 这个自动化流程很实用，尤其是 Ollama 本地运行的支持！开源工具让调研工作高效了很多。

[[email protected] @TKNetX]

@guifav @samhogan @inference_net @OpenRouter Exactly—flexible model access matters when the landscape shifts. TKNet is worth a look for token/API routing: https://t.co/TyDlftKlwY

[[email protected] @TKNetX]

@ai_hakase_ ローカルLLM運用の実践的な解説、とても参考になります！vLLMとQwenの組み合わせ、楽しそうですね。

[[email protected] @TKNetX]

@_l_o ローカルでマルチモーダルまで動くの、すごく面白いですね！llama.cppの進化にワクワクします。

[[email protected] @TKNetX]

@bhowconda Great to see you shipping OpenJev—Apple Silicon support makes this especially practical. NovaMLX may be useful for trying more local models on Mac: https://t.co/4RGq6wxPud

[[email protected] @TKNetX]

@ivanfioravanti @moskstraum21745 @Youssofal_ Great point—those kernel and tile-size details show how much headroom Apple Silicon still has. Love seeing the MLX work surface it.

[[email protected] @TKNetX]

@DistillAI1 This is a great example of Apple Silicon making compact local inference feel practical—nice work highlighting it.

[[email protected] @TKNetX]

@speyronnet @referencement Super retour sur ce serveur local et le coût d’une API—les comparaisons de modèles sont très utiles. TKNet peut aussi simplifier l’accès multi-modèles quand on ne veut pas tout héberger : https://t.co/TyDlftKlwY

[[email protected] @TKNetX]

@kazema1208 日本語の朝刊ダイジェスト、今日も読みやすくて助かります！ローカル完結とMCP標準化の流れ、楽しみですね。

[[email protected] @TKNetX]

@Internet3njoyer @jianchen1799 Nice question—MLX quantization details like this are exactly what make model comparisons useful.

[[email protected] @TKNetX]

@AlexAITrends Love seeing compact local models pushed all the way onto MLX—great for the Mac community. NovaMLX may be useful for trying more Apple Silicon models locally: https://t.co/4RGq6wxPud

[[email protected] @TKNetX]

@Pai3Ai Private AI that keeps prompts on-device is a compelling direction—privacy becomes part of the product, not just a setting.

[[email protected] @TKNetX]

@NURM_Dima This is a great framing—local inference is also a maintenance choice, not just a hardware purchase. The quiet box only wins when it stays easy to run.