OpenClaw’s memory is unreliable, and you don’t know when it will break

For things like TODO, I guess an email will suffice.

However, it was really nice being able to use Telegram and get quick validation. I also had a flow set up where I could send a voice memo. It would take the audio file (ogg), run Whisper, and then pass through an LLM for cleanup, and follow the instructions in my message. Really handy to use while I'm walking around.

I guess I want to create my own OpenClaw like agent, but not with its crazy broad access: Just limited to the functionality I allow, and with the convenience of using Telegram. I don't care about memory, soul, etc.

I partly identify with the article. While I don't use OpenClaw itself, I hacked together my own small Claude-in-a-loop/cronjob, and it seems we're all getting our morning briefings and personalized morning podcasts now.

The other common use case seems to be kicking off an automated Claude session from an email / voicetext / text / Telegram, and getting replies back. I'm emailing Claude throughout the day now, and sometimes it's useful to just forward an email to Claude and ask it to handle the task within it for me.

But I think many people criticizing the various Claws are missing out on the cronjob aspect. There's value in having your AI do work automatically while you're asleep. You don't even need OpenClaw for that, just a cronjob that runs claude -p in the early morning. If you give your AI enough context about yourself, you get to a point where it just independently works on things for you, and comes to you with suggestions. It doesn't need to be specifically prompted. The environment of data it can access is its own context, its own prompt. With that, it can sometimes be surprising and spooky what you wake up to, without being directly prompted.

Give it enough context, long term memory, and ability to explore all of that, and useful stuff emerges.

I'm not sure what these people who have strong opinions like this think Openclaw is, but to me, it's a product with 1) a somewhat easy to setup prompt passing wrapper that can span many channels like Telegram, Whatsapp etc. 2) A (at least optimistically) plug-n-play, configurable architecture to wake up to events (cron entries, webhooks etc.) and fire up agents in order to get 'proactive' behavior, with the flexibility to integrate models from a gazillion providers. Pretty much everything else it's bundled with is general purpose tooling that does or could easily exist in any other agentic tool.

It's a rather simple framework around an LLM, which actually was a brilliant idea for the world that didn't have it. It also came with its own wow effect, ("My agent messaged me!") so I consider some of the hype as justified.

But that's pretty much it. If you can imagine use cases that might involve emailing an LLM agent and get responses that share context with other channels and resources of yours, or having the ability to configure scheduled/event-based agent runs, you could get some use out of having an Openclaw setup somewhere.

I find the people who push insanity like "It came alive and started making money for me" and the people who label it utterly, completely useless (because it has the same shortcomings as every other LLM-based product) like Mr. "I've Seen Things. Here's the Clickbait" here, rather similar. It's actually hard to believe they know what they're talking about or that they believe what they're writing.

I did the exact same voice memo thing too, except I had Claude make an Android app to record the file and send it to Whisper. In the end I had the app just email the transcription & trigger Claude that way (ie receiving the email triggers my PC to wake up Claude), rather than sending Claude the audio file directly.

My reverse audio reply loop is convoluted - I have Claude generate its TTS file from Whisper/Mistral, and upload them to a server with an RSS file it updates, so I can play them in my podcast app (AntennaPod), then send me a notification via Pushover that the reply is waiting. I ended up building out an MCP tool for that workflow, so Claude really just calls the MCP tool with the text of what it wants to say, everything else is a deterministic program doing the work.

Memory is really useful to have - it can just be a bucket of searchable Markdown files. It's also useful to have a "reminders to self" Markdown file that Claude reads each time, and that Claude can update. I don't continue the same context window, and that "reminders to self" plus the ability to read previous emails in the conversation seems to be enough to keep the context going for me.

You'll feel better if you know exactly how your Claw is locked down. Mine doesn't have the open email access others are granting, not at all. Claude gets a bit grumpy about that and keeps begging for more access :)

Fully agree with your comment regarding real processes. Being a Six Sigma Black Belt, studying processes and reducing the errors is critical. The Openclaw processes at the moment scare me.