This is an experiment, not a recommendation - I want to flag that up front. I’ve been using OpenClaw for a couple of days as a mobile front end for personal todos, and I’m writing this down mostly to have a record of what it costs and whether it’s actually solving the problem I think it’s solving.

To be clear: this is for personal todos only - things like errands, house admin, personal reading, that sort of thing. While the value here could be significant, the security and confidentiality needed for work means this type of approach, at least for now, is out of the question.

The Problem It’s Meant to Solve

My todos live in org-mode. Org-mode lives in Emacs. Emacs lives on my desktop. That’s a good system when I’m at the desktop - capturing a task is a keybinding away, and everything is one plain-text file that every other tool on my machine can already read.

The problem is my phone. There isn’t a good org-mode client for mobile that doesn’t involve either a dedicated app (like Orgzly), a sync dance, or typing structured org syntax on a touch keyboard. Most of my “I should add that to my todo list” moments happen away from the desk - on a walk, cooking, half-asleep. That friction means things don’t get captured, which defeats the point of having a system at all.

The Setup

OpenClaw sits on top of this rather than replacing it. Emacs and org-mode remain the actual database - the source of truth is still plain-text .org files in my Dropbox, exactly as before. OpenClaw doesn’t have its own todo list or its own storage. It’s an AI assistant that I can contact by texting it on my phone. When I tell it something needs doing, it appends a properly formatted entry to the right org file over Dropbox.

That distinction matters to me. I’m not migrating my todos into a new proprietary app - I’m giving the existing plain-text system a better mobile write path. If OpenClaw disappears tomorrow, my todos are unaffected; they’re just org files, same as always.

Cost

Setup was about $5 - all API credits to get the assistant working end to end. Ongoing use has settled at roughly $0.10 per query, which for something used a handful of times a day is trivial.

I started on Sonnet for the underlying model, but have since switched to Haiku. For “parse this rambling voice-to-text into a clean org TODO with the right tags and a deadline if one was mentioned” the smaller model is plenty, and it’s meaningfully cheaper per call. I’ll write more if I find cases where Haiku’s parsing goes wrong and I need to fall back to something bigger.

The next obvious step down is dropping hosted models altogether. Parsing a short voice note into a tagged org entry isn’t a task that needs frontier capability - it’s plausibly well within reach of a free, locally-hosted model via Ollama. I already run a server for other projects, so the marginal cost of adding this would be whatever extra power draw the model adds, not a new line item. Haven’t done it yet, but it’s the direction this is heading if the accuracy holds up.

Email, Separately

Worth noting this doesn’t touch an email server at all, but it does pull todos from emails. Email sync and checking happens on-device, since I sync my mailbox to work with Emacs (mu4e) - the same setup I wrote up a couple of months ago. That’s a deliberate separation - OpenClaw’s job, for the moment, is capture-to-org, not handling actual emails.

Early, Unresolved Questions

  • Does the friction saving actually change my capture behaviour, or do I just have the same number of todos arriving via a slightly nicer path?
  • Does Haiku’s parsing hold up as the todos get more complex (multiple deadlines, nested subtasks, scheduling conflicts with existing entries)?
  • Is $0.10/query the honest steady-state cost, or does it creep up once I start using it for more than quick captures?

I don’t have answers yet - this is only a couple of days in. Treating it as an experiment for now and will follow up once I’ve got a longer run of data on whether it’s actually changed anything.