An Agentic Approach to Note and Task Management
Unified Note and Task Management
Previously, I used Obsidian for my knowledge base and Super Productivity for task management. While both were effective, they functioned as separate systems, with information and actionable tasks residing in different locations.
The Search for Convergence
The friction of switching contexts and the lack of visibility between tasks and notes eventually became a bottleneck. I began experimenting with alternatives to bridge this gap. I spent time testing Silverbullet, looking for that perfect balance of networked thought and task execution.
However, my goal wasn't just to find a new tool, but to find a way to unify these streams into a single source of truth - one that was accessible not just to me, but to the AI agents I've been including into my workflow as of late.
The solution arrived with TaskNotes, an Obsidian plugin. By bringing task management directly into my existing vault, I eliminated the need for external extensions and separate toolsets.
The Architecture: PARA and TaskNotes
My current system is built on the principle of simplicity and structured organization using the PARA method. Everything lives within a single directory structure:
- 0-Inbox: For rapid capture.
- 1-Projects: Active endeavors with a defined end.
- 2-Areas: Ongoing responsibilities.
- 3-Resources: Interests and reference material.
- 4-Archives: Completed or inactive items.
I use a combination of Obsidian, Neovim, and Zed to interact with these files, treating my notes as raw, editable text. Task management is now handled via a dedicated "TaskNotes" directory located at the root of my vault. This structural change was pivotal. My tasks are now just markdown files within my vault; part of the same ecosystem as my knowledge base.
Enabling Agentic Workflows
The power of this workflow becomes apparent when I use a command line agent/harness. My current choice for local models is Pi agent.
By using community-made skills for interacting with the TaskNotes MCP, the AI agent can navigate my tasks and notes seamlessly. With a well-defined AGENTS.md file, I can instruct local model, such as Gemma 4 26B, to help me create tasks from the inbox and notes. This new workflow is exceptionally efficient due to the agent having access to my entire vault of valuable context. I've also found use cases such as organizing projects, summarizing progress, creating tasknote-specific "views", and even suggesting next steps based on my task list.
Because the data is unified, the agent doesn't need access to any external tools; it simply needs to read my directory. This creates a loop where the agent assists in managing the very information it uses to understand my context.
Infrastructure and Future Vision
To ensure this workflow is available wherever I am, I use Syncthing, in combination with a vpn, to keep my vault synchronized across my devices. Whether I'm on my desktop or a mobile device, the state of my tasks and notes remains consistent.
My ultimate goal with this setup is replaceability. I've design every layer of my stack so that any single service can be swapped out without collapsing the entire ecosystem. Each layer is intentionally stripped down to the basics - taking inspiration from the Unix philosphy for software desgin. Instead of building a dependency on a software, I'm building a system based on standardized markdown and protocol-driven access.
The next step in my exploration will involve integrating Hermes agent, further pushing the level of autonomy granted to my personal agents.