openvort/openvort

Open-source AI Employee Platform — Create AI employees that collaborate with your real team in WeCom, DingTalk, and Feishu, just like hiring a few more colleagues.

What it solves

OpenVort provides a platform to create and manage "AI employees" that integrate directly into enterprise messaging platforms like WeCom, DingTalk, and Feishu. It solves the problem of disconnected AI tools by allowing AI agents to perform real-world tasks—such as writing code, reviewing pull requests, running CI/CD builds, and generating reports—alongside human teams.

How it works

It uses an agentic loop powered by LLMs (like Claude) that can autonomously decide when and how to call specific tools. The system is built on a plugin architecture where "AI employees" are assigned specific roles and skills. These agents operate within isolated Docker sandboxes for tasks like coding or script execution. The platform includes a web dashboard for management and exposes all tools via the Model Context Protocol (MCP) so they can be used by external clients like Cursor or Claude Desktop.

Who it’s for

It is designed for development teams and enterprises that want to automate repetitive workflows by embedding AI agents directly into their existing communication and development infrastructure.

Highlights

  • Multi-Platform Integration: Works with WeCom, DingTalk, Feishu, and other messaging apps via a unified gateway.
  • Autonomous Agent Runtime: Features an agentic loop that supports tool use, task scheduling, and automatic failover between models.
  • Plugin-Based Extensibility: Includes a marketplace for installing and publishing custom skills and plugins via pip or zip bundles.
  • Isolated Execution: Creates dedicated Docker environments for each AI employee to safely run code, scripts, and build tools.
  • Developer-Friendly: Provides a comprehensive CLI, REST/WebSocket APIs, and built-in support for RAG, Jenkins, and browser automation.

Related

  • Project
    moss-site/moss-trade-bot-skills

    Moss is an open‑source framework that turns natural‑language descriptions of crypto trading styles into fully‑featured quantitative bots. It includes a local backtesting engine, an automated weekly evolution loop that refines tactical parameters, safety guardrails, and optional integration with the Moss platform for verification and live trading. Install via a tagged release, let the AI infer indicators and risk settings, run evolution backtests, and (if desired) bind the bot to the Moss service for real‑time execution.

  • Project
    eddyzzl/marvis-risk-agent

    A local-first, governed credit-risk Agent platform that transforms natural language requests into audited risk analysis, models, and strategies.

  • Project
    BlockRunAI/blockrun-mcp

    BlockRun MCP is an open‑source Node/TypeScript MCP server that gives Claude‑compatible agents 19 real‑time tools—including market data and the ability to place USDC‑settled bets on Polymarket—charged per call via a self‑custody Solana/Base wallet (x402) or a BlockRun API‑key with prepaid credit.

  • Project
    Nutlope/receipthero

    ReceiptHero is an open-source receipt management app that uses AI-powered OCR (via Together AI's Qwen3.5 vision model) to extract structured data from receipt images, categorize expenses, and display a spending breakdown. All data stays in your browser for privacy. It's built with Next.js, Tailwind CSS, shadcn/ui, and Zod, and you can run it locally by setting up a Together AI API key.