BradGroux/veritas-kanban

What it solves

Veritas Kanban provides a structured, local-first task management system that bridges the gap between simple visual boards and complex autonomous AI agent workflows. It allows users to start with a standard Kanban board and incrementally add governance, automation, and agent orchestration only as their needs grow.

How it works

At its core, the project uses local files (Markdown for tasks, JSON for settings, YAML for workflows) to maintain a source of truth. Users can interact with the board via a web interface, CLI, or REST API. For advanced users, the system supports optional layers like the Model Context Protocol (MCP), agent orchestration, and a workflow engine that functions similarly to CI/CD pipelines for AI agents. It includes a policy engine to enforce security, audit trails for agent decisions, and sandbox presets to control what agents can access on your machine.

Who it’s for

It is designed for developers and power users who want to manage tasks locally while experimenting with or deploying autonomous AI agents. It is particularly suited for those who need to maintain strict control over agent behavior, audit their actions, and ensure human-in-the-loop oversight for code-related tasks.

Highlights

• Local-First Storage: Uses human-readable Markdown and YAML files by default, with optional SQLite support for multi-user environments.
• Agent Governance: Includes a policy engine, sandbox presets, and decision audit trails to safely manage autonomous agents.
• Workflow Engine: Enables multi-step, version-controlled agent pipelines with support for human-in-the-loop approval gates.
• Flexible Orchestration: Supports multi-agent teams, real-time dashboards, and shared communication channels (Squad Chat) for task coordination.
• Git-Native: Features isolated worktrees per task, built-in code review, and direct GitHub PR integration.