ObsidianMap — an “AI Chief of Staff” for your Obsidian vault

How I got here

I've been exploring personal knowledge management systems since 2020. For most of that time, my home was Logseq, and I taught a course on it. Obsidian is newer for me — I only moved across at the start of 2026, when Claude Code arrived and changed what I thought was possible.

For the first time, I could put an AI assistant inside the same folder my notes live in, and have it do real work. Read the files. Follow conventions I'd set. Run workflows I'd designed. The vault stopped being a place I stored things and started being a place I worked. That shift is the reason ObsidianMap exists.

The move from Logseq was a real shift, not a cosmetic one. A different model, different defaults, a different way of thinking about how notes link together. But once it settled, the thing I'd been quietly chipping at for years became buildable in a way it hadn't been before.

I suspect this might be one of the last big products I make in this space. Not because there's nothing left to learn, but because this one resolves a cluster of problems that have kept me tinkering for half a decade. When something works, it's worth saying so.

The problem with most Obsidian setups

You've used Obsidian for a year or two. The vault has grown. There are folders inside folders, three competing daily-note templates, half-finished MOCs, and a backlog of ideas you haven't looked at since the weekend you set them up.

The structure isn't the problem. The constraint is the daily ritual. Without one place to start the day and one place that tells you what's actually moving, every morning begins with a small archaeological dig.

What ObsidianMap is

At the centre of ObsidianMap are three layers, for the three horizons you actually live in:

• Compass— orientation. The place you turn to when you need to know where to go. It doesn't hold your notes; it points to them. Think of it as a queryable checkpoint for the whole vault.
• Terrain— what's currently on the cards. The active playing field, across every domain you're working in. Not a filed list of projects, but a view of what actually has movement right now.
• Horizon— the long view. Who you want to be, what you're building toward, the things the other two layers are ultimately in service of. Slower cadence, and ideally kept current by what flows up from the work below rather than by a compilation step you have to remember.

Around those three layers sits the rest of the system: themed backlogs that the daily work pulls from rather than replaces, domain folders for the areas of your life that have substance, and a journal that catches raw thinking before anything gets routed anywhere.

The AI assistant part

Call it a Chief of Staff, call it a co-pilot, call it a thoughtful assistant — the framing matters less than the architecture. What I'm really after is a working knowledge system alongside the vault, designed to be read and written to by an AI as a first citizen, not an afterthought.

I've come to think about that system in four parts:

• Pipelines— the connectors that pull raw material from external systems into your inputs. Think of this layer as your MCP servers: a Fathom script that fetches new transcripts, a Google Calendar puller, a Slack export handler. Today it's Fathom; tomorrow it could be Linear, Jira, or whatever else you're working in. Built to be expandable — you add a pipeline per tool you actually use, no more.
• Inputs— the channels by which raw stuff lands in your vault. Meetings, voice notes, documents, Slack exports, ideas. Each input type has a folder it arrives in and a lifecycle for what happens once it's handled (raw → _processed/, logged in _log.md). The set of channels is open-ended — add a new type when a new source of raw material shows up.
• Skills— the repeatable workflows the assistant runs once raw inputs are in. Named procedures with a clear input, a clear output, and opinions about how to get from one to the other: process-meetings turns a transcript into a structured summary linked to the right project; process-actionstriages the running open-actions and waiting-on lists. Not vague “help me with my notes” prompts — actual procedures the system knows how to execute.
• Memory, context that holds up— the memory layer that keeps the assistant grounded across sessions. The !Memory/files (one per domain), the naming conventions, the section structures, the project status vocabulary. This is the most important layer: it's what stops you from re-explaining your team every Tuesday, and stops the assistant from inventing answers, because the conventions tell it where to look. It's meant to be updated, not static — the system gets sharper the longer you work in it.

A lot of AI-and-Obsidian setups stop at “point an LLM at the vault and see what happens.” That works for a little while, and then begins to fray, because neither the vault nor the prompts were really designed for it. What I'm building is the other end of that — the structure and the workflows shaped together, so the assistant has somewhere solid to stand.

Who it's for

People who already use Obsidian (or want to) and have hit the wall where the vault has become harder to maintain than it is helpful. People juggling several domains — a job, a side project, personal writing, family logistics — without wanting four apps to do it. People who want to use AI in their thinking without handing it the wheel.

Not so much for the “10x your productivity” crowd. More for the person who'd like their Sunday evening to feel a little less like catching up on the last six days.

Where this is right now

I'm using ObsidianMap daily and refining it as I go. The template, the assistant prompts, and the writeup explaining the choices behind each piece will be released together. If you'd like a note when it's ready, drop your email here.