I built Markplane because my AI coding assistant had no idea what I was working on.
I use Claude Code daily. It's good at writing code, but it has zero awareness of project state — what's in progress, what's blocked, what the plan even is. So I'd re-explain context every session, or it'd make decisions that conflicted with work elsewhere in the project, or duplicate things already in progress.
This didn't start as a product. I built it for myself from my own evolving file-based approach to project planning, task tracking, and feeding context to AI assistants in a way that actually worked. Over time I refined it, and eventually realized it was worth formalizing into a proper tool and workflow.
Markplane stores tasks, epics, and plans as markdown files in your repo. markplane init, and your project plan lives next to your code, version-controlled with it. Branch your feature? Your project plan branches too.
In practice, the workflow looks like: I'm talking through an approach with the AI, I say "create a task for this," and later I tell it "implement the auth refactor task" — and it picks it up with full awareness of everything else in the project. The AI isn't just writing code on demand; it's carrying tasks from idea through implementation.
That works because of a .context/ directory that maintains a compressed representation of your full project state. An AI assistant reads the compressed context and understands what exists, what's in progress, and what's blocked — without stuffing hundreds of task files into its context window. Think of it as a materialized view of your project, optimized for LLM consumption.
There's a built-in MCP server so AI assistants can query and update tasks naturally. "What's blocking the auth refactor?" or "create a task to refactor the payment module" just works.
No SaaS, no database, no signup. Single binary, v0.1.2.
I'm most curious about two things:
(1) the .context/ approach — have you run into the problem of your AI tools having no awareness of your project state, or burning through the context window trying to understand enough to work effectively?
(2) How are you handling the gap between "AI writes code" and "AI helps manage the project"?
I built Markplane because my AI coding assistant had no idea what I was working on.
I use Claude Code daily. It's good at writing code, but it has zero awareness of project state — what's in progress, what's blocked, what the plan even is. So I'd re-explain context every session, or it'd make decisions that conflicted with work elsewhere in the project, or duplicate things already in progress.
This didn't start as a product. I built it for myself from my own evolving file-based approach to project planning, task tracking, and feeding context to AI assistants in a way that actually worked. Over time I refined it, and eventually realized it was worth formalizing into a proper tool and workflow.
Markplane stores tasks, epics, and plans as markdown files in your repo. markplane init, and your project plan lives next to your code, version-controlled with it. Branch your feature? Your project plan branches too.
In practice, the workflow looks like: I'm talking through an approach with the AI, I say "create a task for this," and later I tell it "implement the auth refactor task" — and it picks it up with full awareness of everything else in the project. The AI isn't just writing code on demand; it's carrying tasks from idea through implementation. That works because of a .context/ directory that maintains a compressed representation of your full project state. An AI assistant reads the compressed context and understands what exists, what's in progress, and what's blocked — without stuffing hundreds of task files into its context window. Think of it as a materialized view of your project, optimized for LLM consumption.
There's a built-in MCP server so AI assistants can query and update tasks naturally. "What's blocking the auth refactor?" or "create a task to refactor the payment module" just works.
No SaaS, no database, no signup. Single binary, v0.1.2.
https://github.com/zerowand01/markplane
I'm most curious about two things: (1) the .context/ approach — have you run into the problem of your AI tools having no awareness of your project state, or burning through the context window trying to understand enough to work effectively? (2) How are you handling the gap between "AI writes code" and "AI helps manage the project"?