--- title: "Todo — An agent-driven kanban built for solo devs" description: "I published Todo, a local, lightweight project management tool designed for AI agents. Why I built it, and how I use it with Claude and Lain." date: 2026-04-06T14:00:00+02:00 tags: ["Todo", "Tooling", "AI", "Open Source"] --- I've been using Azure DevOps for years to manage my projects. It's a great tool — when you're working in a team with sprints, pipelines, and boards tied to repos. But when you're a solo dev who just wants to **track project progress with an AI agent**, it's overkill. Same goes for GitHub Projects: it works, but there's friction — especially when you want an autonomous agent to create tickets, move them around, and comment, without heavy configuration. So I did what I've been doing a lot lately: I built the tool myself. **→ [Todo on GitHub](https://github.com/Ekioo/Todo)** --- ## The problem When I work on Aekan or other projects, I've gotten used to delegating tasks to **Claude** and **Lain**. I give them a goal, they plan, execute, and report back. But for this to work well, you need a shared tracking tool — a place where the agent can: - See existing tickets and their status - Create new tickets when it identifies work - Move a ticket from one column to another when it's done - Comment to explain what it did (or what's blocking) - Check if it's been mentioned somewhere With Azure DevOps, this is possible via the API, but it requires OAuth setup, permissions, field mapping. With GitHub Projects, you have to go through GraphQL, manage tokens, and the data model is sometimes counterintuitive. Nothing insurmountable, but every time it's time spent on plumbing rather than the project. --- ## The solution: a local SaaS in two days If I can ship a working tool in two days, why not do it? Todo was born from this idea: a kanban that's **custom-built, lightweight, local, and designed from the start to be driven by agents**. In practice, it's: - A classic **kanban board** with columns, tickets, labels, priorities - A **full REST API** under `/api` — everything the UI does, the API does too - A **mention system** with `@agent:claude` or `@owner` to communicate between humans and agents - **Markdown everywhere** — descriptions, comments, cross-references with `#ticket` - **One SQLite database per project** — total isolation, zero configuration - Everything runs **locally** on `localhost:5230` No cloud, no complex auth, no subscription. Clone, `dotnet run`, done. ![The Todo board in action](/images/blog/todo-board-screenshot.png) --- ## How it works with an agent The API was designed so an agent can figure things out on its own. Here's the typical workflow: 1. **Discovery** — the agent lists projects, columns, members, and available labels 2. **Reading** — it checks the board, filters by priority, looks for tickets that concern it 3. **Action** — it creates a ticket, moves it, adds a comment, changes the priority 4. **Communication** — it mentions `@owner` to flag a blocker, or checks its own mentions to see if there's pending work Agents identify themselves with the `agent:{name}` format — for example `agent:claude` or `agent:lain`. This makes it easy to track who did what in the activity history. --- ## A concrete example with Aekan On Aekan, I have a Todo project with columns **Backlog → In Progress → Review → Done**. When I want to move forward on a feature: 1. I create a ticket (or ask Claude to do it) with the requirement description 2. Claude picks up the ticket, moves it to "In Progress", and starts working 3. When it's done, it moves to "Review" with a comment explaining what it did 4. I review, validate (or comment), and move to "Done" If Lain identifies a bug during automated testing, it creates a ticket directly with the context, priority, and mentions `@owner`. I see it the next time I glance at the board. Simple, fluid, no ceremony. In reality, on Aekan, it's not one or two agents — it's **13 specialized agents** running autonomously, orchestrated by a custom engine I call the **dispatcher**. It handles routing, parallelism, learning, and agent evaluation — all wired into Todo's API. I go into detail in [a dedicated article](/blog/dispatcher-agent-orchestration). --- ## Why open source This tool solves **my** problem, but I'm convinced it can help others. If you work with AI agents and you're looking for a lightweight tracking tool with no friction that you can evolve however you want — that's exactly what it was built for. The code is clean, the stack is simple (.NET 10, Blazor, SQLite), and the API is documented via OpenAPI. Fork it, adapt it, contribute. **→ [github.com/Ekioo/Todo](https://github.com/Ekioo/Todo)** --- ## Follow the development If this kind of tooling interests you — or if you want to see how I use AI daily in my projects — join me on **Discord**. That's where I share in real-time. **→ [Join the Discord](https://discord.gg/4MVPfw9wTQ)**