Project Astra像谷歌深度神经网络那样,在真实时刻内看, 聽, 和思考整个世界的一种通用 AI 代理

5.0 (4)

审阅者 Daniel Nikulshyn·更新 2026年5月

AI Agent Multimodal Voice Assistant Computer Vision Research Prototype Google DeepMind Wearables Real-Time

1 / 3

概览

Project Astra 是 Google DeepMind 的一款实验性通用 AI 助手，旨在通过以人类的方式理解世界来帮助完成日常任务。它可以同时处理视频、音频、图像和文本，让用户通过指向摄像头或自然发声来获得上下文感知的回应。基于 Google 的 Gemini 模型，Astra 被设计为低延迟的对话交互，并且能够持久记忆最近的上下文。它定位为一个研究原型，探索通用代理如何最终在手机、智能眼镜和其他环境设备上运行。虽然尚未公开可用，Astra 预示着谷歌在代理式 AI 方向上的发展，该技术能够观察周围环境、记住所见，并代表用户执行有益行动。

主要功能

实时视频和图像理解
基于语音的会话接口
持久性的上下文存储
文本, 音频和视觉层面的多模态推理
与吉米(Gemini)模型家族集成
支持眼镜, 手机等设备

价格

模型: Freemium
分类: Multimodal AI
评分: 5.0 / 5 (4)

使用场景

通过手机摄像头进行视觉问答

使用手机指向物体, 文字或场景, 并问出语音问题来接收实时的上下文相关解释。使用 Astra 的实时视频和语音理解功能。

智能眼镜上的无手帮助

穿戴适配的智能眼镜来获得周围的会话式辅助信息, 并利用 Astra 的低延迟的多模态推理。

日常任务的上下文记忆

询问参考 session 早期时刻的跟进问题,例如回忆最后在哪里看到一个物品, 使用 Astra 的持久性短期记忆。

基于吉米的通用多模态代理的研究探索

使用 Astra 作为原型, 来学习如何通过吉米模型在真实时刻内构建的通用性多模态代理, 来观察, 推理, 和回应

优点 & 缺点

优点

实时多模态理解
自然,低延迟的语音会话
深度神经网络研究支持
上下文和短期记忆在 Session 中
为可穿戴和移动设备设计

缺点

对公众尚未普遍开放
限制详细的数据处理
尚在实验性原型阶段
能力可能因设备而异

评测

5.0

4 个评分的平均值。

登录以留下评测。

Gunnar Eriksson

Jan 18, 2026

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on persistent contextual memory, and context and short-term memory across a session caught me off guard. still, I'd recommend giving it a real trial.

Aisha Khan

Jan 6, 2026

Use it every day

Honestly didn't expect to like it this much. Voice-based conversational interface is exactly what I needed, and designed for wearables and mobile devices. but I reach for it almost every day now and it just clicks.

Linda Petersen

Aug 14, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is voice-based conversational interface — handled better than most — and real-time multimodal understanding. Not broadly available to the public is my one real gripe. Worth the time if this is your use case.

Rina Desai

Jul 2, 2025

Compared a few options

Evaluated this against two competitors. Where it wins: integration with Gemini model family and designed for wearables and mobile devices. Where it lags: limited details on data handling. On balance the feature set — especially prototype support for smart glasses and phones — justifies the 5 stars for our use case.

问答

Is Project Astra available to the public, and how can I access it?

No, Project Astra is currently an experimental research prototype from Google DeepMind and is not broadly available as a public product. Google has demonstrated it publicly but has not released general access details.

What can Project Astra actually do with video, audio, and images?

Astra performs real-time multimodal reasoning across text, audio, images, and live video. Users can point a camera or speak naturally and get context-aware responses, with persistent short-term memory letting it recall what it has recently seen or heard within a session.

What devices is Project Astra designed to run on?

Astra is being prototyped for phones, smart glasses, and other ambient or wearable devices. However, capabilities may vary across devices, and full device support has not been finalized since it remains a research prototype.