At Medevel.com, we always round up open-source and AI projects in drafts, primarily for us to review each one, install it, test it, find alternatives, and write about it. However, as the number of open-source and AI projects scales every day, thanks to talented developers who utilize AI to solve real problems and create awesome open-source apps that aim to replace expensive commercial software, we encounter dozens of new solutions daily.
These tools often provide capabilities that we previously believed required immense resources and manpower to produce.
Therefore, we have decided to curate a monthly roundup of the best open-source and open AI projects to help our audience find what they need. Note that these lists will not be categorized; they will contain a variety of apps and solutions across different categories and purposes, serving both daily users and engineers or advanced skilled users, which has been the goal of Medevel.com from the start. To make it easier, we will define what each app does in its title.
The list is ordered randomly based on our findings for the month. We have excluded several apps and projects that are unfinished or lack proper documentation. We owe it to you, the users, the developers, and the open-source community to find the best tools and provide you with the most usable apps in these roundups.
So, here we go.
Concat Your Free CapCut Alternative
Concat is a free and open-source (GNU Affero GPL 3.0) app that many may consider is the true CapCut alternative. It is built to run on macOS, Windows, and Linux. It does not require any subscription and best yet, comes with dozens of useful tools that we used to see only in highly expensive and commercial apps.
One of the best features that we liked about Concat is, it comes with built-in text-to-speech, voice cloning, dozens of premium effects, text editing and advanced keyframe control. It exports to several formats such as 264, HEVC, AV1. Up to 4K 60, 10-bit colour.
Concat is built with Rust, so expect seamless performance as advertised.
Diagram Design Github
Diagram Design is a MIT-licensed skill for AI agents that generates clean, editorial-quality HTML and SVG diagrams without relying on Mermaid or complex shadows.
It supports diverse layout grammars like Sankey and UML, ensuring semantic clarity and accessibility while allowing seamless conversion of existing draw.io or Excalidraw sources into polished, self-contained visuals.
FlyDrones: Neural Drone Autopilot
FlyDrones simulates a fruit fly connectome in your browser to pilot a drone, translating 850 neurons and optic flow into real-time flight controls without human input.
Stickman video director
Stickman Video Director is a Codex Skill that transforms text ideas into high-quality animated stickman videos with confirmed voiceovers and Gemini Omni Flash prompts.
It supports customizable durations and multiple visual styles, from minimalist black-and-white to modern character designs, optimized for YouTube Shorts, TikTok, and educational content.
The tool streamlines production by converting scripts directly into ready-to-render prompt packages, enabling fast creation of engaging visual storytelling without complex animation skills.
Gods Eye: Ghost Edition (Monitoring)
Gods Eye Ghost Edition is a free, keyless browser-based 3D globe for tracking live aircraft and ships. It eliminates paid API dependencies like Google Maps or OpenAI, offering immediate zero-cost access to satellite visuals and public data feeds without accounts or credit cards.
OpenStock: Real-time Stock Monitor
OpenStock is an open-source alternative to expensive market platforms. Track real-time prices, watch the whole market, and explore detailed company insights, built openly, for everyone, forever free.
God’s Eye View
God’s Eye View is an open-source, browser-based spy-satellite simulator rendering a photorealistic 3D globe with live aircraft, ships, and seismic data.
Controlled by a hands-free AI agent, it merges public signals into an inspectable, real-time cockpit experience. Run it locally without API keys to track the world, break the code, and see everything.
System One Har: Typed Decision Loops (AI DEV)
System One Harness transforms static decision models into autonomous agent loops by replacing fragile text generation with typed, confidence-gated actions.
It executes one precise model call per step, ensuring every transition is probabilistic, auditable, and traceable for robust, crash-proof autonomy.
WorldClaw: Agentic 3D Open World Creation
WorldClaw transforms single open-ended prompts into explicit, explorable 3D worlds by using agentic planning to ensure global terrain coherence while generating rich, locally detailed environments.
It produces high-quality scenes where every object and terrain feature remains a separate, editable instance, allowing for seamless modification and exploration across diverse biomes like snowline villages or tropical islands.
MyClaw3D: Visualize your AI Agent in 3D Workspace
MyClaw3D is an isometric 3D workspace built with React and Three.js that visualizes AI agents performing local animations while connecting to real-time chat via an OpenClaw gateway.
Helios: Real Real-Time Long Video Generation Model
Helios is a breakthrough 14B video model that generates high-quality, minute-long videos at nearly 20 FPS on a single H100 GPU.
It achieves this speed and coherence without relying on standard anti-drifting tricks or acceleration techniques, outperforming smaller models in both quality and efficiency.
Jev Social: Automated Social Media Analysis
Jev Social is a local-first research agent that automates Instagram, TikTok, and LinkedIn analysis by routing decisions through Jev while capturing cited browser evidence via the socai CLI.
Maic: The AI Apple Silicon LLM Server
Maic is a high-performance, MLX-optimized local LLM server for Apple Silicon Macs. It provides an OpenAI-compatible API and modern web interface, enabling fully offline, privacy-first inference without subscriptions or API keys.
XGEN-JING: An Egocentric
Interactive Experience Model
XGEN-JING is an egocentric interactive model that generates synchronized first-person video and audio for navigation, object interaction, and conversation based on user actions and reference images.
Built on MiniMax-H3, it enables immersive exploration of virtual environments through keyboard-controlled movement and text-guided dialogue with consistent audiovisual output.
Project Lyra: Open Generative 3D World Models
Project Lyra is NVIDIA’s open-source series of generative 3D world models, evolving from single-image reconstruction in Lyra 1.0 to explorable, long-horizon 3D environments in Lyra 2.0.
Hypit: Clone Viral Videos
Hypit empowers AI agents to clone viral videos by replicating entire workflows, not just scripts. Generate hundreds of variants with swapped faces, B-roll, and captions in a single command, enabling rapid content scaling without relying on expensive generation models. It is free & open-source.
G0DM0D3: Multi-model Chat for Red Teams, and Researchers (AI)
G0DM0D3 is a fully open-source, privacy-transparent, multi-model chat interface that pushes the limits of the post-training layer, for red teaming, cognition research, and liberated AI interaction. Built for hackers, philosophers, and system tinkerers.
AgentMemory
AgentMemory is the memory solution that can saves you time, money and resources on your AI Agent. It offers a persistent memory for Claude Code, Codex CLI, Gemini CLI, OpenCode and more.
three.ws Give your AI Agent an Avatar
three.ws is an open-source, browser-native platform for creating and embedding 3D AI agents. It generates rigged GLB avatars from text or images, equips them with LLM brains, memory, and wallets, and enables seamless web integration via a single component.
It supports on-chain identity and x402 payments, it offers a complete toolkit for building interactive, autonomous 3D personas without server dependencies.
Hermes3D: A 3D Workspace for AI Agents
Hermes3D is a self-hosted 3D virtual office that visualizes AI agents collaborating in real-time, replacing static dashboards with an immersive, interactive workspace for managing your automated workforce.
mlx-workspace: Local Coding Model Benchmarker for Apple Silicon
mlx-workspace benchmarks local LLMs on Apple Silicon via MLX to determine which models can reliably handle routine coding tasks without cloud API costs.
viral2viral – UGC Advertisement Video Generator
viral2viral is a free open-source app that analyzes successful UGC ads to extract key visual styles, messaging tones, and engagement techniques using AI. It then generates new promotional videos for your products that replicate these proven strategies to maximize marketing impact.
OpenShorts.app: Free AI Podcast Clipper
OpenShorts is a self-hostable, MIT-licensed AI platform that automatically generates vertical clips from podcasts using smart stacking and face-tracking without configuration.
It offers a free CPU-based local option or a faster hosted service, providing flexible tools for creators to repurpose long-form content into engaging shorts.
WorldGrow: Generating Infinite 3D World
WorldGrow is a hierarchical generative framework that creates infinite, photorealistic 3D worlds by synthesizing structured scene blocks through context-aware inpainting and coarse-to-fine strategies.
It overcomes the scalability limits of existing methods to deliver unbounded, geometrically coherent environments suitable for next-generation interactive world simulators.
Run llama.cpp on 2-core CPU + 8GB DDR2: My Full CPU-only Guide (4B models at ~2 t/s)
An awesome guide by 0ut0flin3, tutorial on how to run llama.cpp on a low hardware specs machine. You can review the full gist here. Running CPU-only inference for 2B-4B parameter models is highly viable on legacy hardware like a Core 2 Duo with 8GB RAM by using lightweight Linux distributions and optimized llama.cpp configurations.
0ut0flin3′ approach delivers fluid, crash-free performance that allows for simultaneous multitasking, proving that efficient local AI doesn’t require modern GPUs or extensive resources.
