What would a community edition of Vibe Kanban look like?

[stunningpixels]

Hello Community,

As we approach two months since announcing that the startup behind Vibe Kanban will be shutting down, I wanted to start a discussion about whether and how we might reorganise the project so it can continue as a fully open-source, community-maintained edition.

The key questions to understand are:

• What does Vibe Kanban solve today that other tools don't?
• What should be added to the roadmap?
• Who is willing to donate time to maintain the project?

Louis

[stunningpixels]

To kick off the discussion, some thoughts:

• VK has always been two things: short/medium term project management tooling and a daily coding agent orchestrator
• Daily coding agent orchestrators from the labs, especially Codex Desktop, have gotten really good and are being updated extremely frequently
• I don't get excited about a roadmap that's essentially: feature catchup with extremely well staffed labs
• Perhaps a way around this is to think about how VK could be used with other daily coding agent orchestrators

[ericnunes30]

Having a project management agent would be interesting, especially with the heartbeat functionality.

[stunningpixels]

I think a project management agent might just be a coding agent. The evidence for this:

• Linear are pivoting into building a coding agent
• OpenAI shipped agent thread management just this week
• We built an 'orchestrator' agent earlier in the year (unreleased) and tested the concept, partial implementation of the MCP part here

[JoshuaDietz]

My main usecase for Vibekanban is working on the same project (or different projects) with multiple Opencode agents in parallel without them influencing each other. So basically a browser based Opencode-orchestrator. From a feature perspective the tool covers my usecase almost entirely. So I don't care about project management features at all, for me this is a tool for professional developers to orchestrate agents in an efficient and clean way, sometimes even from the phone. My main problem are the bugs. With a little more polish this tool has the potential to become orders of magnitude more helpful than it already is. Some of the bugs that hurt my workflow the most are:

• subagents get stuck often (and that I cannot peek into them like I can in Opencode makes it worse)
• question tool often not showing the question
• unable to copy larger portions of text from the logs view because when scrolling, the text out of the visible area gets removed
• weird markdown formatting when pasting error messages into the chat box (in my opinion just raw text would be enough)
• cannot paste images on first message in new session
• diff view performance problems for longer diffs (scrolling getting stuck)
• "running" button state seems to be glued to the session instead of to the worktree, so to restart my backend I have to switch back to the old session that I started it with
• hitting "stop" often makes a page reload necessary
• queuing messages does not seem to be reliable (chat getting stuck, message stays queued forever)

And from a feature perspective I am mainly missing small features:

• easy way to switch between modes using the keyboard (currently have to use the dropdown either by mouse or my keyboard, I'd love something like being able to tab through agents like switching between plan and build mode in opencode)
• being able to jump between questions in the question tool
• being able to see the chat of subagents
• llm generated titles/names for chats and branches instead of just using the beginning of the message
• llm generated names for sessions
• ability to give a "context brief" for a worktree that is automatically injected into each new session
• more efficient way to choose repos when creating a new worktree (currently have to click through multiple menus)
• ability to merge/rebase all repos from the chat at the same time
• ability to resize the different panels (i.e. the diff view)
• ability to change the diff view between "all on the worktree" and "since last commit (or even a specific one)". Sometimes when I let an agent iterate, it would be so helpful to be able to see what exactly was changed since my feedback instead of when the worktree was started.
• alternative way to define the URL that is opened in the browser panel. In some of my configurations I have to start multiple processes and the first URL that gets output is not necessarily the one for the frontend.
• ability to start a new chat (or session) with comments in the codebase (even in unchanged areas). This way it would be possible to start a refactor by looking at code, commenting it and then letting the agent run on these comments.

Bigger features /innovation stuff would be for me:

• ability to run agents isolated in containers/VMs (but Vibekanban running on the host, worktrees being synced) - this would be a big one for me. Currently I run all of Vibekanban in a VM which has some annoying drawbacks
• integration with process-compose so vibekanban can automatically start databases and other dependencies (but only once)
• Projects, but in another way than it was before. For me, a project would be an entity that groups together all chats that belong to the project, a project has 1..n repos and might have additional context info (i.e. snippets that can be added to the prompt easily), possibly even dependencies (database, object storage, ...) and a recipe on how to start them (globally, for the whole project) and maybe a "machine" (the docker or vm abstraction from above) on which this project should be ran. When starting a new chat, that'd be done in a project. I can then choose 1..n of the repos that belong to the project.

What I personally don't care about at all (but I'm aware that others might do)

• kanban features
• project management features
• github integration (I use another git provider)
• other orchestrators than Opencode

In terms of time to contribute: I was planning to do it since a few months, but realistically speaking I will not be able to find the time to do so in the coming months :/ when it comes to money, I'd be willing to contribute a few euros per month (Patreon or similar)

Final thoughts: Vibe Kanban is such a help tool for my workflow. My main show-stopper right now are the bugs listed above, especially the state management bugs where agents are getting stuck (or appear to get stuck)

[2001zhaozhao]

I'm working on an unreleased agent orchestrator tool called Kotkit that implements basically the complete feature-set from Vibe Kanban while solving a lot of its problems. (Plus add a lot of features like a desktop windowing interface, built-in VSCode/terminal emulator/file explorer, and highly flexible plugin support, and is much faster than VK.)

Out of your problems with VK, I realized that I already solve quite a lot of them. (tbh this is meant for a list for my own reference later as much as it is an ad for my unreleased tool):

Details

• Subagents stuck (on the UI side, I have some subagent UI differentiation, but they're mostly still shown as tool calls on the main agent UI just like VK)
• Question tool often not showing the question
• Unable to copy larger portion of text from the log view
• Cannot paste images on first message in a new session (my tool unifies attachment system between kanban system and agent system. It otherwise works like vibe-kanban's)
• Hitting "stop" often makes a page reload necessary (my tool is server-rendered which eliminates whole categories of issues like this one)
• Queuing messages does not seem to be reliable (my tool has a simpler, more reliable queue system with server-side queued message storage)

Issues in your list that my tool might still have:

Details

• Weird markdown formatting when pasting error messages into the chat box (I copied a lot of the text editor setup of vibe-kanban. Can't find text that reproduces this bug right now but I've definitely ran into it with VK and it may also be in my app.)
• "Running" button state seems to be glued to the session instead of to the worktree (in my experience these issues are sticky and hard to debug, i don't think i got all of them yet)
• Diff view performance problems for longer diffs (i haven't implemented diffs yet. when i do, i will pick a very scalable, high performance engine)

Feature requests you made that I do have already:

Details

• Being able to jump between questions in the question tool
• More efficient way to choose repos when creating a new workspace (in Kotkit you can pre-pick set of repos, and customize a default repo list for new workspaces in a project. 99% of the time i never have to customize repos at all in a new workspace)
• Ability to merge/rebase all repos from the chat at the same time
• Ability to resize the different panels (i.e. the diff view) (some sidebars were actually already resizable in VK, but in my tool all of them are)
• Integration with process-compose so vibekanban can automatically start databases and other dependencies (Kotkit is a single web server that uses SQLite and nothing else so it works out of the box)

Features you mentioned that i think are good ideas (or are already planned):

Details

• Being able to see the chat of subagents (note: VK already renders them for agents that show subagent logs, but does so separately from the main agent. In Kotkit, i could add workarounds based on reading persisted session files even for the agent types that don't expose subagent logs natively.)
• LLM generated titles/names for chats and branches instead of just using the beginning of the message (I plan to introduce a new "automated task" category of agent settings that you can customize different from the normal settings for things like this, with different categories of tasks you can give to different models. For example running opencode on a specific small model ID with very small system prompt and limited toolset.)
• LLM generated names for sessions (same as above)
• Ability to give a "context" brief for a worktree that is automatically injected into each new session (i plan to do this with AGENTS.md/CLAUDE.md creation at the root of a multi-repo workspace)
• Ability to change the diff view between "all on the worktree" and "since last commit (or even a specific one)". (I haven't gotten to implementing diff view yet because i want to get its optimization right, but i will definitely have this feature when implementing it.)
• Alternative way to define the URL that is opened in the browser panel. (I will have a comprehensive Run Configuration system inspired by IntelliJ, as well as auto port assignment so you can open multiple dev servers without conflict. For remote development of web apps, the Kotkit server proxies the web traffic so inside address bar you can just type localhost regardless of where you're accessing Kotkit from.)
• Projects as entities that group together all chats that belong to the project, containing repos each with additional context info and dependencies. (Already partly implemented with specific repo-set per project. Planned Run Configuration support and per-project coding agent context injection support via project-root-level AGENTS.md as stated above)

Features you suggested which i probably won't add to my tool soon:

Details

• Easy way to switch between models using the keyboard (I will have a keybinding system but i'm not sure how detailed)
• Ability to start a new chat (or session) with comments in the codebase (in Kotkit, every agent has kanban board access by default, so you're supposed to ask the previous task's agent to create a ticket for the next one to carry over context. i don't think i see a good use case for using comments in the code instead.)
• Ability to run agents isolated in containers/VMs (this isn't soon but planned much further down the line)

[JoshuaDietz]

Sounds interesting, is an open source release planned for the future?

[2001zhaozhao]

Yes, it will be released as open source sometime in the next few months, maybe in this month.

[MartyLake]

What I liked in the first editions (before the cloud stuff was introduced) :

• unified project management (kanban) interface for different aspects of the project
• solves the problem of: which of the X coding agent terminal instances was the one where I wanted to solve Y aspect of the project
• solves the problem of: I am coming back to the project after a break, where was I ? what was I trying to achieve ?

[hugochinchilla]

I don't know how much time I can donate to this but I would like to contribute for sure. I'm currently maintaining my own branch with some minor patches that fix small annoyances .

When I started working on it I realized there is a need for adding some testing, most code is untested.

Some of the features listed above are really good ideas, but personally I would start by removing stuff. The new task management system is too complex, tries to solve what most of us already solve using other tools.

But most importantly we need to find a way to workaround the new situation with claude -p and it's pricing

[2001zhaozhao]

I don't know about vibe-kanban, but in my tool I will release several different workarounds to address the claude -p pricing change.

[JoshuaDietz]

The only proper solution around the Claude pricing changes at this point is not using Claude. Whatever will be done, we would always enter a cat and mouse game that willleave users annoyed, accounts being locked and time wasted. It's obvious that Anthropic only wants to offer their highly subsidized subscription to users willing to use their harness directly.

Yesy it's unfortunate because the Claude models are the best, but other models are quite usable as well. I think the project is better off by focusing on them and not on weird cheats just to somehow get Claude working for a few weeks until they get patched.

[CydeSwype]

I think Linear has moved into a similar space recently with project management views on coding agents and it's a good model to describe what a problem to solve looks like. It's also reflecting what Louis said in his "what would you do differently" response: get big corporate clients (I'm paraphrasing). I'm specifically trying to solve the kanban view of projects as an interface to coding agents AND handling multiple clients (like... people who pay me money for my work). I have three partners/clients and each had provisioned me coding agent tokens (mostly Claude) and being able to easily plan and execute parallel work for each is a real problem for me that I think will be more common in the future as more people take on more contract work with these tools.

[soabwahott]

A good start would be to revert most of the changes made in 0.1.44

[dominikj-cf]

Project management may be overlapping with Linear and other tools but I still find it adds some value in selecting and adjusting tasks intended to be handled by an agent. Although a better integration to 3rd party task trackers might be nice.

For personal projects, VK is enough for me end-to-end.

That said, I think that the primary value of Vibe Kanban for me is the

• worktree management
• good hook defaults (commit)
• prompt presets
• good PR visibility
• single place to use and locate my agent sessions across providers (Claude, Codex).
• Automation primitives (MCP to run/evaluate workspaces)
• Full windows/mac/mobile support. Easy to expose my local node via Tailscale without additional servers, no lock-in

Some of these are solved by other tools, such as Conductor, but less so in the open-source space. Also besides a few bugs, I found Vibe Kanban can scale quite well and works reliably.

I'm willing to put some time into the development, already put up a few PRs, although I'm not sure how aligned they are with the eventual roadmap.

[hugochinchilla]

I think Vibe-Kanban should leave project management behind, too many competitors with much more resources are filling that space. It's trivial for each user to connect a MCP server to their preferred task management suite and create a skill so that claude manages the task status and sync.

I personally have a skill that for each PR created scans linear, finds the matching task or creates a new one, assigns it to me and connects it to the github PR. Then when the PR is merged the linear task is closed. This can be replicated for any project management suite and keeps VK focused on doing only one thing and excelling at it.

[hugochinchilla]

A incredible feature would be to allow a workspace to create new workspaces, sometimes I work on creating plans that span multiple PRs, it would be incredible to write in the prompt "create a new workspace for each task" and have vibe kanban open the new workspaces with a prompt for each defined task.

[dominikj-cf]

VK already supports this via the built-in Vibe Kanban MCP. It is great.

[diykorey]

Hi everyone - I run a small software company in Ukraine, and we'd be happy to help maintain this project for a modest fee. We could prioritize issues and feature requests, and for critical work, the community could pool a small amount to cover development costs. Would something like this be worth exploring? I see