Plumy

Live / Open Source

Local-first project planning for teams

Open-source planning software for teams that want Timeline, Kanban, and Roadmap planning surfaces, optional AI-assisted workflows, and a local-first desktop path without accounts, hidden telemetry, or a hosted planning service.

How It Works

Plan visually

Timeline planning gives teams a date-based view of project swimlanes, delivery windows, workload shape, and upcoming conflicts.

Useful when you need to understand sequence, duration, and conflicts before work moves into execution.

Execute in Kanban

Kanban columns keep the same tasks moving through execution, review, and handoff without duplicating planning data.

Status, owner, notes, comments, and agent handoff state stay attached to the same task.

Architecture overview

Plumy grew from a visual planner into a local-first workspace with desktop persistence, backup portability, release packaging, Roadmap planning, and an optional MCP interface for agent workflows.

01

Desktop shell

Electron owns the desktop runtime, packaging, file attachment handling, external links, local storage access, and the local MCP server lifecycle.

02

Renderer workspace

The React app keeps the planning UI focused on one shared workspace model, with Timeline, Kanban, Roadmap, people, preferences, and details panels reading from the same state.

03

Three planning surfaces

Tasks can be scheduled visually on a timeline, moved through status columns in Kanban, or grouped into roadmap milestones without splitting the data into separate tools.

04

Local persistence

The app is local-first. Electron store acts as the canonical desktop persistence layer, while renderer storage remains useful for portability and backup-friendly flows.

05

Workspace portability

Backup and import include tasks, comments, people, projects, swimlanes, status columns, preferences, MCP settings, and UI state rather than just a partial task export.

06

Agent interface

A local MCP surface exposes workspace snapshots, task and card projections, board polling, structured comments, and gated revision-protected handoff flows on the user device.

Engineering challenges

The difficult parts were less about making another task board and more about keeping planning, execution, persistence, backups, and agent access aligned around one durable workspace model.

Keeping Timeline, Kanban, and Roadmap in sync

Plumy has to let the same task live comfortably in date-based planning, status-based execution, and milestone planning. The challenge is preserving one source of truth while each view feels natural.

Building dense drag-and-drop planning

The timeline needs swimlanes, horizontal scrolling, task positioning, resizing, reorder flows, and collision-resistant interactions without making the workspace feel heavy.

Making local-first feel dependable

A local desktop product still needs recovery, migration, backup, import, separate dev and packaged stores, and clear storage diagnostics so users can trust the data model.

Designing safe agent workflows

The MCP layer is useful only if it stays understandable and controlled. Plumy exposes native planning primitives to AI systems while keeping execution local, reviewable, and revision-checked when agent access is enabled.

Planning system

Plumy is built around the idea that planning and execution should not require separate tools or duplicated task records.

  • Timeline view supports calendar-like scheduling across project swimlanes.
  • Kanban view turns the same work into status columns with persistent filters for execution and prioritization.
  • Roadmap view adds milestones, task links, filters, and explicit dependency arrows for longer-range planning.
  • A shared WorkspaceReadModel feeds Timeline, Kanban, Roadmap, task details, milestone details, and dialogs.
  • People management separates human teammates from agentic teammates and tracks workload context.
  • Task details support markdown notes, structured comments, and concise completion summaries.
  • Preferences expose MCP diagnostics, backup/import, storage usage, and audit log surfaces.

Agent workflow layer

  • Plumy is one of the few open-source projects I have seen so far that offers a credible paid-tier style AI interaction model while still running on a local device.
  • Agents are not treated as a chat overlay. They can inspect the full workspace snapshot plus task, Kanban-card, and Timeline-card projections through stable MCP tools.
  • Board watchers can poll specific statuses with persisted duplicate suppression, so AI systems can participate in real planning loops instead of one-off prompt exchanges.
  • Agents can add structured comments and activity entries, update summaries, assign work, and move tasks through review-oriented statuses when write capability is enabled.
  • Write operations use expected revisions and can run in read-only or gated modes, giving local AI workflows a product-grade control model instead of unrestricted automation.

Product layer

The public Plumy site emphasizes open-source, local-first planning. Under the surface, the product also needed packaging, release automation, docs, a web migration path, and a planning model that can be projected into multiple product surfaces.

  • The public site positions Plumy as open source, local-first, no-account planning software.
  • The desktop app is packaged for macOS, Windows, and Linux through release automation.
  • The GitHub Pages site is built from the same project and deployed separately from the desktop renderer.
  • Plumy Web now extends the product with Convex-backed data, Clerk account flows, roadmap support, comments, and assigned work.

Key Features

  • Timeline planning across project and people swimlanes
  • Kanban execution with persisted filters and flexible status columns
  • Roadmap planning with milestones, task links, and dependency arrows
  • Human and agentic teammate management
  • Markdown task details, comments, and structured completion notes
  • Local MCP workflows with paid-tier interaction quality for AI systems
  • Cross-platform desktop packaging for macOS, Windows, and Linux

Technology Stack

ElectronReactTypeScriptViteTailwind CSSreact-dndelectron-storeMCP

The story of Plumy

Plumy started from the need for a planning tool that could show the shape of work without forcing every team into a hosted project-management system. The important idea was combining Timeline, Kanban, and Roadmap planning around the same task data, so planning and execution could stay connected.

As I kept building it, the scope expanded from a simple visual planner into a desktop workspace with local persistence, workspace backup, people management, richer task notes, roadmap milestones, and agent-friendly project context. The MCP response confirms that this is a live operational workspace rather than a static demo: agents can read task projections, watch boards, and participate in structured review handoffs when access is enabled.

Discovered use cases

The strongest use cases are the ones where planning needs both a calendar-shaped view and a status-shaped view:

  • Planning project timelines while keeping execution visible in Kanban.
  • Grouping longer initiatives into roadmap milestones with dependency context.
  • Using swimlanes to group work by projects, initiatives, or people.
  • Keeping notes, comments, and task details close to the actual work.
  • Backing up an entire local workspace instead of exporting only individual tasks.
  • Letting agents inspect assigned work, poll status boards, and leave structured handoff notes for human review.
  • Testing how local-first tools can support AI workflows without becoming cloud-first by default.

Some fun facts

3

synchronized planning surfaces: Timeline, Kanban, and Roadmap

3

desktop release targets: macOS, Windows, and Linux

10

focused tests across hooks, task utilities, workspace services, and MCP

24k+

source lines across the desktop app, Electron services, and Pages site

147

TypeScript, TSX, CJS, and MJS source files in the desktop project

100+

live workspace tasks visible through the MCP snapshot during dogfooding

Biggest challenge faced

The hardest part was keeping multiple projections of the same workspace consistent: Timeline, Kanban, Roadmap, task details, milestone details, people, backups, and the MCP API all needed to agree about what a task is and how it moves through work.

Code

Plumy combines product design, frontend engineering, desktop app packaging, local persistence, and agent workflow design. The UI and architecture were shaped around a practical question: can a local planning app become a useful coordination surface for both people and AI assistants?

What's next for Plumy?

The plan is to keep refining the desktop product while using the web version to explore collaboration, hosted sync, and account-based workflows.

  • Continue hardening Plumy Web collaboration on top of the Convex and Clerk foundation while preserving the local-first desktop story.
  • Improve MCP workflows with clearer diagnostics, safer remote access guidance, and stronger review handoff conventions.
  • Refine timeline ergonomics for dense project plans, including load visibility, scrolling, and task placement.
  • Expand Roadmap dependency validation, circular dependency warnings, and milestone reporting.
  • Expand backup, restore, and portability workflows so moving a workspace remains predictable.
  • Keep simplifying the product language so Plumy stays understandable to non-technical teams while still supporting advanced AI workflows.

I enjoyed building Plumy because it sits at the intersection of product operations, planning UX, and agent workflows. It is practical software, but it also lets me explore how tools should change when AI assistants become part of the working team.