Vex
An open-source GitOps skill version manager designed for AI agents, operating as a headless cognitive tool. It provides agents with versionable and rollback-capable skill context management via the MCP protocol, and supports automatic ingestion of code repositories through GitHub Webhooks.
Timeline (last 90 days)
Vex is an open-source project that uses the GitOps version control paradigm to manage AI Agent context and skills, with the core concept of treating context as immutable, versionable skills.
Vex is defined as a headless cognitive tool that runs as a background service, supporting Docker Compose and local bare-metal processes.
Vex automatically ingests code repositories by configuring GitHub Webhooks, and agents query the required context in real time via MCP.
Vex reads conventional commits (such as feat: and fix:) and custom operational commits (such as roll: and branch:), and based on these automatically branches, updates, or rolls back the agent's memory state.
Vex is built on MCP and claims out-of-the-box support for Claude Desktop, Cursor, and any MCP-compatible client.
Vex uses Tree-sitter to logically parse and chunk code rather than simply splitting by character count or line count, thereby preserving semantic structure.
Vex's technology stack uses SQLite for the metadata queue, Qdrant for dense vector handling to support semantic retrieval, and a FastAPI + Huey combination for concurrent task processing.
The author's local test results show that the FastAPI + Huey asynchronous architecture can handle 500 concurrent GitHub push loads without SQLite locking issues, and real-time latency remains within 300ms when simulating 50 concurrent agent reads.
The above performance data are all self-test results from the author's local environment and have not yet been verified by third parties in a real production environment.
Vex's future roadmap includes introducing Rust as a chunking engine for more advanced sub-chunking, adding global GraphRAG to improve cross-file associative retrieval, and supporting more programming languages.
1 more timeline events
All Facts (11)
The Vex project has been open-sourced on GitHub, with the repository at Shuuida/Vex-Skillgit.
50%UnverifiedVex is defined as a headless cognitive tool that runs as a background service, supporting Docker Compose and local bare-metal processes.
50%UnverifiedVex automatically ingests code repositories by configuring GitHub Webhooks, and agents query the required context in real time via MCP.
50%UnverifiedVex reads conventional commits (such as feat: and fix:) and custom operational commits (such as roll: and branch:), and based on these automatically branches, updates, or rolls back the agent's memory state.
50%UnverifiedVex is built on MCP and claims out-of-the-box support for Claude Desktop, Cursor, and any MCP-compatible client.
50%UnverifiedVex uses Tree-sitter to logically parse and chunk code rather than simply splitting by character count or line count, thereby preserving semantic structure.
50%UnverifiedVex's technology stack uses SQLite for the metadata queue, Qdrant for dense vector handling to support semantic retrieval, and a FastAPI + Huey combination for concurrent task processing.
50%UnverifiedThe author's local test results show that the FastAPI + Huey asynchronous architecture can handle 500 concurrent GitHub push loads without SQLite locking issues, and real-time latency remains within 300ms when simulating 50 concurrent agent reads.
50%UnverifiedThe above performance data are all self-test results from the author's local environment and have not yet been verified by third parties in a real production environment.
50%UnverifiedVex is an open-source project that uses the GitOps version control paradigm to manage AI Agent context and skills, with the core concept of treating context as immutable, versionable skills.
50%UnverifiedVex's future roadmap includes introducing Rust as a chunking engine for more advanced sub-chunking, adding global GraphRAG to improve cross-file associative retrieval, and supporting more programming languages.
50%