• AI Engineering
  • Posts
  • Apple Releases FastVLM: Blazing-Fast Vision-Language Model

Apple Releases FastVLM: Blazing-Fast Vision-Language Model

.. PLUS: Eigent – 100% Open Source Multi-Agent Workforce

Author

Sumanth P
September 24, 2025

In today’s newsletter:

  • Glean:LIVE – AMA + engineer-led demos (Sept 25)

  • Apple’s FastVLM – Blazing-Fast Vision-Language Model

  • Eigent – Local, Open Source Multi-Agent Workforces

Reading time: 3 minutes.

FREE LIVE SESSION — TOGETHER WITH GLEAN

The Glean team is organizing a live technical deep dive with their engineering team on September 25 at 9:30 AM PT.

The session will cover:

  • Internals of the new Assistant and how it handles personalization at query time

  • How to program agents with “vibe coding” (rapid agent composition without boilerplate)

  • Latency and accuracy benchmarks from real-world workflows

  • Under-the-hood demos of ranking pipelines and orchestration, followed by a live AMA with engineers

If you are exploring large-scale personalization, relevance modeling, or multi-agent workflows, this session provides implementation details and performance results you can apply to your own systems.

Now, let’s get back into the newsletter!

Apple released FastVLM, a vision-language model optimized for speed and on-device performance. Think lightning-fast multimodal reasoning without depending on cloud infrastructure.

Why it matters:

  • Blazing Speed - Hybrid vision encoder (FastViTHD) delivers up to 85× faster response and 7.9× lower TTFT than LLaVA-OneVision

  • On-Device First - Runs locally on laptops and smartphones, boosting privacy and responsiveness

  • Compact + Capable - Smaller model variants make deployment practical without losing performance

  • High-Resolution Support - Processes images up to 1152×1152 pixels for detailed understanding

Apple also shipped WebGPU support and released models on Hugging Face, signaling they want developers to experiment openly and not just on iPhones.

🔗 Check out the Github repo and Hugging Face demo

Imagine building your own specialized workforce of AI agents completely local, fully open source.

Eigent is a desktop application that lets you build, manage, and deploy custom agent workforces to turn complex workflows into automated tasks.

Key Features:

  • Multi-Agent by Default – Developer, searcher, document analyzer, and multimodal agents work in parallel.

  • Zero Setup – Lightweight desktop app, quick install.

  • Customizable – Add your own models, plugins, or API integrations (Slack, Notion, browsing).

  • Human in the Loop – Requests feedback when ambiguity arises.

  • Prebuilt Templates – Research assistance, travel planning, reporting, and more.

  • MCP Integration – Native support for multi-tool coordination.

  • Drag & Drop Workflows – Visual workflow editor makes automation more intuitive.

This is one of the first fully-local open-source multi-agent solutions. It’s like AutoGPT, but designed as a sustainable desktop platform with transparency and user control.

That’s a Wrap

That’s all for today. Thank you for reading today’s edition. See you in the next issue with more AI Engineering insights.

PS: We curate this AI Engineering content for free, and your support means everything. If you find value in what you read, consider sharing it with a friend or two.

Your feedback is valuable: If there’s a topic you’re stuck on or curious about, reply to this email. We’re building this for you, and your feedback helps shape what we send.

WORK WITH US

Looking to promote your company, product, or service to 150K+ AI developers? Get in touch today by replying to this email.