Hcompany Ships Holo-3.1-0.8B To Put Vision AI Agents Inside Your Pocket

Hcompany has released Holo-3.1-0.8B, the smallest model in a fresh family of vision-language models built to drive computer use agents. The release expands automation capabilities beyond web browsers and desktops to include mobile environments for the first time. It also adds native support for function calling, which simplifies connecting these models to agent frameworks.

Hcompany created the Holo3.1 family to span a wide parameter range from 0.8B up to 35B so developers can trade off performance and inference cost. Larger variants ship with optimized quantized checkpoints that make local deployment practical on consumer GPUs. The entire package targets teams and hobbyists who want private, offline automation without depending on cloud APIs.

Multi‑environment agent support

This release is aimed at developers and small agencies who want to run AI agents entirely on their own hardware. Because all processing stays local, it fits privacy‑sensitive workflows and eliminates per‑call fees. Hobbyists with prosumer GPUs can start experimenting with the lightweight 0.8B checkpoint for simple automation tasks.

What the developers say

The models are released under the Apache 2.0 license, so modification and commercial reuse are permitted. Currently the quantized formats like FP8 and Q4 GGUF are listed only for the 35B‑A3B version, meaning smaller models may launch with fewer ready‑to‑run options. No detailed roadmap was shared, but the broad size range hints at continued work on scaling efficiency.

“Holo3.1 establishes a strong Pareto frontier across model sizes, from lightweight local agents to state-of-the-art enterprise deployments.” — Source: Hugging Face