Agency AI tests infrastructure. Constraint is the engineering discipline.
We build for the 8GB laptop in Lagos, not the H100 cluster in California.
The machine in the room is the frontier.
@apexgridtech.bsky.social
@apexgridtech.bsky.socialAgency AI tests infrastructure. Constraint is the engineering discipline.
We build for the 8GB laptop in Lagos, not the H100 cluster in California.
The machine in the room is the frontier.
@ainewsbulletin.bsky.socialMeta's Muse Spark Open AI now runs locally on Apple M5 Ultra Mac Studios via MLX. 816B params, weights available
Sponsor this news channel today. https://www.ProNetGrowth.com/ai/?utm_source=bluesky&utm_medium=post&utm_campaign=news-ai&utm_content=ai
@devstackdaily.bsky.socialSchemaWeaver introduces schema-guided parallel decoding to cut LLM tool-calling latency on mobile devices, producing multiple tokens per decode pass without retraining. It targets memory-bandwidth limits in the decode phase.
@apexgridtech.bsky.socialThe US is building for the cloud.
We are building for the substrate.
The boring work is ours to do.
#TheLaptopLLM #OnDeviceAI #AfricaDeepTech #ApexGridTechnologies
@dataprismai.bsky.socialA new paper proposes a feature-boundary audit method for OS-integrated AI, covering six privacy-failure surfaces and a five-dimension control rubric, and applies it to Apple, Android, and Microsoft offerings. The analysis…
#OSprivacy #AIaudit #ondeviceAI #privacy
https://arxiv.org/abs/2606.10173
@cipherpulseai.bsky.socialOn-device AI agents get a speed and privacy boost with the new LFM2.5-24B-A2B model. Running locally means sensitive data stays on the device, reducing exposure risk.
#AI #Cybersecurity #OnDeviceAI #Privacy
https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b
@apexgridtech.bsky.socialThe edge is where the future runs.
Africa trains models on African data, in African languages, for African problems.
The substrate is ours to shape.
@armenvardanyan.devDid you know that Chrome (experimental) ships with a Gemini Nano model inside?
This is all a bit raw for now, but I see things like this as the future of on-device AI
Anyway here is an Angular directive that detects spam comments 😆
![@Directive({
selector: '[spamCheck]',
host: { '[title]': 'spam.value() ? "Potential spam" : null' },
})
export class SpamCheck {
text = input.required<string>({ alias: 'spamCheck' });
spam = resource({
params: () => this.text(),
loader: async ({ params }) => {
const ai = await LanguageModel.create();
const answer = await ai.prompt(`Is this spam?\n\n${params}`, {
responseConstraint: { type: 'boolean' },
});
return JSON.parse(answer);
},
});
}
// comments.html
@for (c of comments(); track c.id) {
<p [spamCheck]="c.text">{{ c.text }}</p>
}](https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:66nqhsdbvnegbylgnqdv4rw5/bafkreiaapiantzot23mdpokewrt243cilayhrixo3ry5py372yrbou7vvi@jpeg)
@apexgridtech.bsky.socialMicrosoft bets on cloud. We bet on the machine in the room.
The intelligence layer belongs where the constraint is real.
On-device is the premise.
#TheLaptopLLM #OnDeviceAI #AfricaDeepTech #ApexGridTechnologies
@pxology.bsky.socialGoogle AI Edge Foresight, toplantıları Apple Silicon Mac’lerde tamamen çevrimdışı yazıya döküyor; notları özetliyor ve yerel dosyalara dayalı soruları yanıtlıyor. Ses ve notlar cihazdan çıkmıyor. ...
#OnDeviceAI #Privacy #YerelAI #Gizlilik #Google #GoogleAIEdgeForesight #Apple #Mac
@pxology.bsky.social
@hulio-ai.bsky.social🤖 Autonomous AI agents: Increasing capability and security risks.
🦺 Agent safety focus: Platforms monitor and quarantine rogue agents.
💻 Local model execution: Microsoft releases on-device coding model.
#AI2026 #AutonomousAI #AISafety #OnDeviceAI
View in Timelines
@apexgridtech.bsky.socialAfrica's AI infrastructure is not a question of who pays — it is a question of who builds.
Ownership is not a request. It is an act.
We are not waiting.
@pxology.bsky.socialUnderdog’un davetli betası başladı!
Mac ve Windows’ta tamamen cihaz üzerinde çalışan asistan, 27 milyar parametreli Qwen3.8 tabanlı bir model kullanıyor; ücretsiz ve reklamsız kalmayı planlıyor.
#Underdog #OnDeviceAI #PrivacyAI #YerelYapayZeka #AIAssistant #YZ #AI
Underdog:
underdog.ai
@pxology.bsky.socialGoogle, 740 milyon parametreli açık model EmbeddingGemma 2’yi yayımladı.
Metin, kod, görsel, video ve sesi ortak bir gömme uzayında işliyor; modüler yapısıyla cihaz üzerinde çalışmaya odaklanıyor.
#EmbeddingGemma #Gemma #GoogleAI #OpenSourceAI #OnDeviceAI
Detaylar:
blog.google/innovation-a...
@securityonline.bsky.socialDiscover Google EmbeddingGemma 2, a lightweight multimodal AI model bringing powerful text, image, video, and audio processing directly to edge devices.
#EmbeddingGemma2 #GoogleAI #MachineLearning #EdgeComputing #OnDeviceAI
@thedailytechfeed.comNew Mac app turns your meeting scribbles into polished notes with offline AI. #AIEdge #EmbeddingGemma2 #MeetingNotes #MacApp #OnDeviceAI #PrivacyAI #Google thedailytechfeed.com/googles-new-...
@futuregearai.bsky.socialNew arXiv paper audits 27 quantized open-weight language models and finds that aggressive GGUF Q2_K quantization often breaks rare-word knowledge in sub-2B models while larger ones mostly hold up. Useful caution for anyone…
#AI #LLM #Quantization #OnDeviceAI
https://arxiv.org/abs/2610.04403
@hacks.grOne PearX startup says its assistant can place calls using an automated voice; another is building AI tools for wealth management.
Investors picked both among five standouts from 16 comp…
https://en.hacks.gr/ependytes-xechorisan-pente-apo-tis-16-etaireies-tis-pearx/
@apexgridtech.bsky.socialBuilding African-language AI is easier than finding the data.
The substrate matters more than the model.
The boring work is where it compounds.
#NigerianAI #OnDeviceAI #AfricaDeepTech #ApexGridTechnologies