12 Open-Source AI Projects for Agentic Workflows

the gist

A curated list of open-source tools for AI agents, covering video production, cybersecurity, codebase indexing, and local voice processing.

OpenMontage: A framework that turns AI coding assistants into video production teams by handling scripting, asset generation, and composition using tools like Remotion.
DeerFlow: A ByteDance-developed agent harness designed for long-horizon tasks, utilizing sub-agents, memory, and sandboxes to automate complex workflows like data pipelines and dashboard generation.
GStack: A collection of skills codified by Garry Tan that structures agent behavior into a Y Combinator-style startup process, including phases for planning, CEO reviews, and deployment.
Hermes Agent: A high-star repository focused on self-healing and self-improving agentic workflows, serving as an alternative to existing agent harnesses.

Anthropic Cybersecurity Skills: A set of six cybersecurity frameworks (including MITER and NIST) that can be plugged into any agent-supporting CLI to perform automated security audits on codebases.
Codebase Memory MCP: An indexing engine by Deus Data that indexes large repositories (e.g., the Linux kernel) in minutes, allowing for structural queries in under 1 millisecond with 3D visualization.
SkillSpector: An Nvidia-built security scanner that inspects agent skills for 65 vulnerability patterns, including prompt injection and data exfiltration, before installation.

Hyperframes: A framework for converting HTML, CSS, and animation libraries like Three.js into deterministic MP4 videos, suitable for product demos and motion graphics.
Unlimited OCR: A vision-language model from BYU that performs high-speed document analysis and PDF highlighting, with a model size of approximately 6.5 GB.
Palmier Pro: An open-source, AI-native video editor for macOS that features a built-in MCP server, allowing external agents to control the editing process.
Voicebox: A local voice I/O stack that combines speech generation and transcription, supporting voice cloning and local model integration for a complete audio pipeline.