You NEED to try these 12 open-source AI projects RIGHT NOW
TL;DR
Hermes has crossed 200,000 stars and leans hard into self-healing agents: Berman calls it a strong OpenClaw alternative because it can automatically fix failed skills and improve on the next run.
ByteDance's DeerFlow is built for long-horizon work, not quick prompts: With nearly 74,000 stars, it orchestrates sub-agents, memory, and sandboxes for tasks like building data pipelines, dashboards, slide decks, and content workflows over hours or days.
Open Montage tries to turn an AI coding agent into a video studio: The 15,000-star project can handle research, scripting, asset generation, editing, and final composition across 12 pipelines like trailers, explainers, podcasts, and documentary montages.
Several projects are really agent skill packs, not full apps: Anthropic cybersecurity skills, Matt Pocock's 143,000-star software engineering skills, and Garry Tan's 114,000-star G Stack all package expert workflows so agents can work more like practitioners.
Codebase Memory MCP focuses on speed at codebase scale: Deus Data says it can index the 28 million line Linux kernel in 3 minutes, answer structural queries in under 1 millisecond, support 158 languages, and use 120 times fewer tokens.
The roundup also includes practical media tools, from OCR to local voice to AI video editing: Highlights include Baidu's 6.5 GB vision-language OCR model, Palmier Pro for agent-controlled video editing on macOS, Hyperframes for HTML-to-MP4 rendering, and Voicebox for local speech generation and transcription.
The Breakdown
A GitHub repo with 200,000 stars that fixes its own failed skills, a ByteDance agent harness built for tasks that run for days, and an open-source voice stack that aims to replace both ElevenLabs and Whisper all made Matthew Berman's weekend open-source roundup worth paying attention to.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.