kandev
kandev is a self-hostable Kanban workbench designed specifically for orchestrating coding agents. It provides a visual interface akin to a Kanban board where each task can be assigned to a different AI agent, enabling multi-step workflows with human-in-the-loop gates. This ensures that at each stage—whether it's code generation, testing, or review—a human can approve or reject the output before proceeding, adding a layer of safety and quality control. The system supports parallel named sessions, allowing multiple workflows to run concurrently without interference. Each session is isolated using git worktrees, meaning every agent operates in its own clean checkout of the repository. This prevents conflicts and accidental overwrites, making it safe to run several agents simultaneously on different branches or even across multiple repositories. The multi-repository support extends the tool's utility to complex projects spanning several codebases. kandev is particularly suited for teams or individual developers who want to leverage AI agents for coding tasks but need fine-grained control over the process. By self-hosting, users maintain full data privacy and can customize the environment to their infrastructure. The open-source nature allows for community contributions and transparency. Overall, kandev bridges the gap between autonomous AI coding and human oversight, making it a practical tool for accelerating development while mitigating the risks of fully automated code generation.
- Category
- AI Agents
- Pricing
- Open Source
Problem solved
Managing multiple AI coding agents in a controlled and isolated manner is challenging; without proper orchestration, agents can produce conflicting changes, and fully autonomous workflows risk introducing errors. kandev solves this by providing a Kanban workbench with step-wise human gates, git worktree isolation, and parallel session management, allowing developers to safely harness multiple agents across various repositories while retaining oversight.
Use cases
- Orchestrating sequential AI coding agents for feature development with human review at each step
- Parallel code experiments using multiple AI agents on isolated branches
- Safely running autonomous code refactoring across multiple repos with approval gates
- Onboarding new team members to AI-assisted coding with controlled, step-by-step workflows
- Building a self-hosted AI code generation pipeline with Kanban visibility
Audience
- Developers
- ML Engineers
- Researchers
- Startups
Pros
- Human-in-the-loop gates ensure code quality and safety
- Git worktree isolation prevents conflicts between concurrent agent tasks
- Supports parallel sessions for scaling AI-assisted development
- Multi-repository support accommodates complex project structures
- Open source and self-hosted for full data control and customization
Cons
- Requires infrastructure setup and maintenance for self-hosting
- Learning curve for configuring multi-agent workflows and git worktrees
- Dependent on the underlying coding agents' capabilities
- Limited to coding tasks, not general-purpose agent orchestration
- Community and ecosystem may be nascent
Official website: https://github.com/kdlbs/kandev
Canonical LambdaBase page: https://www.lambdabase.com/tools/kandev