# The Open Source Radar - Week of the 15th December

URL: https://www.onsomble.ai/blog/open-source-radar-dec-15-agentic-ai

Author: Rosh Jayawardena

Published: 2025-12-14T12:00:00+00:00

Updated: 2026-08-02T12:12:41.52786+00:00

We are seeing a shift from AI that chats to AI that acts. This week, we look at 5 open-source projects redefining the build—from autonomous coding agents to infinite video generation.

![Radar display tracking emerging open-source AI tools and technology signals](https://media.onsomble.ai/published/blog/3942a5c1-a03b-439b-903f-8ae88e27b326/hero-001/9bd0d2157335123e77b968f3b2e3c26a4b3c58bc1020f0ad898088d01f24f459/1435.webp)

It’s becoming a bit of a full-time job just keeping up with open source right now. You blink, and the industry has moved on from “chatbots” to full-blown autonomous agents.

There’s a lot of noise — but there are also genuinely impressive tools being built that change how we think about software. I’ve been digging through repos with a simple filter: **things that feel shippable, and meaningfully push the “AI that acts” frontier forward**.

The shift is pretty clear: we’re moving away from AI that just _talks_, towards AI that _does_. Here are five projects that caught my eye this week.

* * *

### 1) ElevenLabs UI

If you’re building voice agents, you’ve probably noticed that a static text box feels a bit dead. As we move towards multimodal interaction — think Gemini Live or OpenAI’s advanced voice mode — the interface needs to catch up.

**ElevenLabs UI** is a neat library of React components that brings voice experiences to life. It gives you polished **orbs and waveforms** that visualise listening, thinking, and speaking states. It’s built on top of **shadcn/ui**, so it drops cleanly into a Next.js app.

It’s a small detail, but it’s the difference between an app feeling like a robot… and feeling like an interface.

**Repo:** [https://github.com/elevenlabs/ui](https://github.com/elevenlabs/ui)

* * *

### 2) Open Computer Use

Anthropic made a splash recently when they gave Claude the ability to control a mouse and keyboard. Naturally, the open-source community took that as a challenge.

**Open Computer Use** is a platform that effectively gives your AI agents _hands_. It orchestrates two roles: a **Planner** that breaks down what you want to do, and an **Executor** that actually clicks buttons and types text.

This is the bridge between generative AI and useful AI. It opens the door to automating workflows across legacy software and web interfaces — not just APIs — so the agent can actually get work done rather than narrate what it _would_ do.

**Repo:** [https://github.com/LLmHub-dev/open-computer-use](https://github.com/LLmHub-dev/open-computer-use)

* * *

### 3) Full Stack Agent

This one is equally impressive and slightly terrifying (for those of us who write code). **Full Stack Agent** is a “text-to-app” platform.

You give it a prompt, and it handles the end-to-end lifecycle: database setup, backend logic, frontend build, and deployment to a live URL. It also runs in a sandboxed environment (which matters a lot when you’re letting an agent generate and execute code).

For rapid prototyping and internal tooling, this is huge. It compresses the “first few hours of a project” into minutes, so you can spend your time on the logic and the user value instead of scaffolding.

**Repo:** [https://github.com/FullAgent/fulling](https://github.com/FullAgent/fulling)

* * *

### 4) Dexter

Most AI agents try to be good at everything and end up being average at best. **Dexter** takes the opposite approach: it’s designed to do one thing really well — financial analysis.

Think of it as an autonomous junior analyst. It pulls in market data, reads financial documents, sanity-checks its own work (crucial in finance), and produces a structured report you can actually use.

It’s a great example of **vertical AI**: constrain the scope, increase the reliability, and suddenly you’ve got something that feels far closer to enterprise-grade. If you love Claude Code, Dexter will feel right at home for you.

**Repo:** [https://github.com/virattt/dexter](https://github.com/virattt/dexter)

* * *

### 5) Stable Video Infinity

Generative video is having a moment, but most tools still struggle with duration — they tap out after 5 or 10 seconds, or drift into visual nonsense.

**Stable Video Infinity** tackles this with a “prompt stream” technique: chaining scenes and transitions to produce longer-form video that aims to keep coherence across the narrative. You can script a sequence that moves from a city street, to a drone shot, to a close-up — and it’s built around the idea of maintaining continuity as the video evolves.

If you’re interested in storytelling rather than just short clips, this one is worth watching.

**Repo:** [https://github.com/vita-epfl/Stable-Video-Infinity](https://github.com/vita-epfl/Stable-Video-Infinity)

* * *

## The takeaway

The common thread here is **agency**.

We aren’t just prompting models for answers anymore — we’re giving them interfaces, tools, and execution environments so they can _operate_ inside real systems and complete real tasks.

If you try any of these, I’d love to hear what you end up building.

## Category

[Engineering](https://www.onsomble.ai/blog/category/engineering)

## Tags

- Agents
- Generative AI
- LLMs
- Open Source
- Transformation
