The main difference is the product direction. Screenpipe seems focused on continuously giving agents context through APIs, MCP, and skills. Daydream is more narrowly built around answering "what did I do today?" through a timeline you can inspect, replay, search, and turn into a daily digest.
I'm also treating deletion as part of the data model. If you cut a sensitive span, its frames, audio, OCR, transcripts, embeddings, and summaries should be deleted or invalidated too.
Mine is still early and Linux-first. I'm open-sourcing it in case anyone wants to contribute, poke around, or use it as a starting point. It’s built with Tauri, a Rust backend, React/TypeScript, SQLite, GStreamer, Whisper, OCR, and VLM processing.
I genuinely didn’t know you were building this when I started. Apparently personal memory capture is becoming a SaaS category too lol.
Code is here: https://github.com/snackbit/daydream
I'm curious what does your AI buddies do more specifically and how do you use them?
We capture everything and gives you and option for a flexible data retention policy, so that you can prune mp4 files for example or verbose accessibility trees, while our agents generate high level memories referencing low level data
It still not perfect, but ideally Screenpipe would record forever so that you have infinite memory of every tiny details, while having high level structured memories for humans and agents, individually or as a company
in my experience, even within a single project, the more agentic cycles, the more features and the more surface area the more cruft builds up. This is an obvious observation, but I suppose I’m at the point I think the move forward is to create a new project, keep only what worked from the preceding as modularized entities and intentionally wipe away all past of how we got there save for the API interface truths.
During normal usage, how often does it try to parse info from the screen capture? Once a minute?
Screenpipe captures accessibility tree and screenshot when you perform a meaningful action. It doesn't use AI at recording time (except PII removal and OCR infrequently). So it can be every few hundred milliseconds to few minutes or more (if screen idle or sleeping)
We usually benchmark CPU usage on $200 Windows/MacOS laptop
That's not just specific to digital stuff. E.g. warehouse cameras that film anything else beyond securing entries/exits, like they are commonplace in e.g. the US don't exist for the same reason.
We also have early support for encryption at rest through our CLI
Ultimately, it’s up to you what you run and trust
Screenpipe is open source and local-first
You can inspect it and keep your data on-device, with encryption at rest
you're being unnecessarily hostile about a threat that almost certainly doesn't exist. And if it does, you can absolutely block those uploads before they even happen. In fact if you don't already block uploads to hosts you don't know about, then you're not really taking anything seriously, are you?
if this doesn't run fully local its a no go for enterprise let alone ordinary users
https://screenpipe.com /how-to-install
I think that sucks, but it's certainly not what I thought of when I read your one-liner (which implies that they harvest and abuse your actual recorded data).
Have you ever founded a company and tried to find paying customers?
You'd probably be offended if I sent you a cute 90's era AOL CD with our software on it.
OP: I wouldn't be offended if you emailed me in this way. Ignore the naysayers. This is perfectly kosher.
In fact, I'm going to check out your company in depth just because of this pearl clutching.
Keep building and don't let naysayers tear you down over something not even slightly evil. (Venomous attacks are more evil than this, frankly.)
For one, even taking the ethos of local-only as something bulletproof, I can't think of a pro that would outweigh the con of recording everything I do. Part of that might be my lack of imagination, as I'm not a heavy user of ai besides basic Claude code.
Most important though, the concept is ripe for pivoting away from local-only. It will only take a juicy investment offer and I'm sure local-onpy will be quietly forgotten.
We're the only local-first, open source option
How much time do you spend explaining the context to your Claude? Or writing detailed prompts? Copy pasting documents into the prompt?
>How much time do you spend explaining the context to your Claude? Or writing detailed prompts? Copy pasting documents into the prompt?
Prompts: not much. If I feel my prompt is becoming a paragraph-long text, that's when I know I probably haven't thought things through enough yet. Documents: I typically have a interpreter console open where I'll do something like File.write("foo.html", page.body), and I'll just tell Claude to "find the missing div in foo.html". Or if its a file already in the repo, I won't need to copy paste anything. Context: also not much. It has all the context by default, doesn't it?