Vektor: the task API my AI assistants run on

I'm building both of them like products, even though right now the user is mostly me. They're past the sketch stage. They're not launched. The front doors are public. Everything behind the sign-in stays there.

A to-do workspace is lists of tasks. A project workspace is projects, and a project can hold notes, files, links, passwords, and its own lists. Tasks move. Search covers the whole account, which is more useful than it sounds once you've been using it for a month.

There's an API because I got tired of pasting tasks into a box for the agent. A key is shown once and then only a hash is stored. The key opens what I granted it: one workspace, one project, or one list. It can be read-only, allowed to mark things done and leave notes, or allowed to actually change things. Making a new project or a new list is a separate switch. I want a key that can update a task without also being allowed to invent a project I didn't ask for.

It can ping another system when something changes. Those pings don't include passwords. A key that can open a project can still see the passwords in that project, so I keep those grants small on purpose.

Sign-in is at calebdorris.com/apps/vektor.

Most task apps add an API later. I built Vektor the other way around. Assistants that watch my inbox and calendar, a weekly planning routine, and Cursor sessions in a repo all record work as it happens. If each kept its own notes, I'd have a dozen lists that disagree. So they all read and write one.

What it exposes

• Workspaces, projects, and lists, on a to-do side and a project side.
• Tasks with status, due dates, sub-tasks, notes, and images. A task can move lists and keep its notes.
• Project notes, files and folders, and reference links.
• One search across everything a key can see.
• Signed webhooks for a workspace or a single project.
• Errors that name the record and the reason, so an agent can fix its own request.

The docs are part of the API

The manual is an endpoint. It comes back as Markdown, changelog on top, with a short section on how to work. Every assistant reads it before touching anything, so a feature I ship today gets used next session.

curl -H "Authorization: Bearer vek_YOUR_KEY" \
  https://example.com/vektor/api/v1/docs

House rules

Search before creating. Anything that belongs to a bigger effort goes in a project, and an agent can start one when it needs to. One-offs stay on a to-do list. When work moves, update the same task and leave a note. That's how a six-week SEO plan for this site ended up as a file in its own project. On October 23, my work workspace alone held 10 projects and 150+ open tasks, and I can still find things.

A key per agent

Each key is fenced to one workspace, one project, or one list. It's read-only, done-and-notes, or full write, and creating projects is a separate switch. One key only opens a single venture project. When an assistant tried to make a new project with it, Vektor said no.

I take the cash accounts I've marked as included. If the bank sent an available balance, I use that. If it didn't, I use the current one. Then I subtract the bills due before my next payday, and I subtract whatever I've parked in a vault. That's the number on the home screen.

Cards and loans show up, because I want to see the debt, but they don't get subtracted again. A paycheck that hasn't landed yet shows up too, and it doesn't get added in. I'm not in the business of spending money I don't have yet. No payday on the account means Assay just looks thirty days out.

It has accounts, a plan, and an admin, the way a small service does. Nobody is being charged. The only plan is called Founding and it's zero dollars. Banks can connect through Plaid once I turn that on. Until then I can keep a ledger by hand and the number still works.

The front door is calebdorris.com/apps/assay.