Wanted to share something I built that might be useful here.
Informity AI is a free, open source Mac app that indexes your local documents and answers questions from them — entirely on-device. No cloud, no accounts, no telemetry. After initial setup it works with no internet connection at all.
Full stack runs locally: model inference, embeddings, vector search, chat history. Web search is opt-in (off by default, requires your own API key, documents never involved even then).
MIT licensed, source is on GitHub if anyone wants to verify the privacy claims.
https://www.informity.ai | GitHub - informity/informity-ai: Informity AI · GitHub
Happy to answer questions about the implementation or how to verify what network calls it makes (short answer: none, except optional web search with your key).
5 Likes
Is this a personal project to show proof of concept for such a tool?
Also, will you also be updating the release with newer models with updates? Is this product still in beta or is stable?
2 Likes
I hope you can also support EPUB files. It’d be really helpful for all students to use their e-textbooks to answer for questions faster.
I also really want to see this for be made available for desktop Linux.
2 Likes
Not a proof of concept at all — this is a fully functional, stable app that I use myself every day. Personal project yes, but built to actually work, not just demonstrate an idea.
And yes, models will keep getting updated. I started with Qwen3.5 35B and switched to Qwen3.6 35B as soon as it came out. When better models drop, I update. That’s one of the advantages of local — you’re not waiting on someone else to decide when you get access to something newer.
Actively adding features too, so more coming.
3 Likes
You’re in luck — EPUB support just landed in v0.9.9. You can grab it at https://www.informity.ai (Apple Silicon).
On Linux: I actually built a .deb package for Debian/Ubuntu, but I only have my MacBook Pro M so I can only build for arm64 right now. To support x86_64 I’d need access to an Intel machine to build and test on. Working on it.
2 Likes
Thank you so much for getting back. That’s great to hear.
A .rpm would be preferred because Fedora feels like a more robust distro that’s gaining in popularity at the same time. So, please consider it as well.
Is there a reason why Flatpak here is not what you’re going with?
2 Likes
With any new tool, app, project, the common concern is the longevity of support from the dev.
Can you comment on that as well? I hope you won’t give up on the project anytime soon.
2 Likes
Answering both in one go:
On RPM and Flatpak — Mac came first simply because that’s what I have and use daily. Apple Silicon’s unified memory architecture is what makes running a 35B model practical on consumer hardware without a discrete GPU, so it was the natural starting point. Linux is next and you’re right that Flatpak is probably the smarter approach — one package that works across distros rather than maintaining separate .deb and .rpm builds. I’ll look into it seriously.
On longevity — fair concern with any indie project. Honest answer: this started as something I built for myself and grew from there. The roadmap is extensive enough that I won’t be running out of things to build anytime soon. I can’t promise enterprise-level support but I’m genuinely invested in this and plan to keep developing it. MIT license also means the code lives on regardless of what happens.
3 Likes
This looks brilliant! Exactly what I’ve been looking for. Thanks so much for building this. Does it (or are there any plans to) support Apple Pages, Numbers, etc… documents?
2 Likes
Thank you, really glad it’s useful!
Numbers should be fairly straightforward to add — standard parsing libraries handle it well.
Pages is a bit trickier under the hood. It’s actually a package rather than a flat file, so it needs some pre-processing before indexing. Totally doable though, I’d handle that automatically so you wouldn’t need to do anything differently. Just point at your folder as usual.
I’ll add both to the roadmap, thanks for the nudge!
2 Likes
Amazing, thank you. A huge number of the documents I use for work are in Pages so that would be very useful.
2 Likes
Informity AI v0.11.0 is live: smoother chats, clearer progress, and stronger startup reliability
Hey everyone,
v0.11.0 is now out, and this release is all about making Informity AI feel more reliable and less confusing in everyday use. You can download this latest version at https://www.informity.ai/ as always.
What’s better in this release
- More reliable responses while chatting
- I fixed cases where a response could look short while streaming, then appear fuller after refresh.
- Answers now stay in sync more consistently in real time.
- Clearer “it’s still working” feedback
- Long replies now show better progress cues so it’s obvious the app is still actively generating.
- Status updates are cleaner and reset correctly between phases (searching vs generating).
- Fewer setup headaches
- If a saved local model setting becomes invalid, Informity AI now auto-recovers to an installed model instead of dropping you into a confusing setup loop.
- Improved response quality consistency
- I removed an internal post-processing behavior that could sometimes over-shape responses.
- Result: answers are now more direct and natural to what the model generated.
Why this matters
This release is focused on trust:
- Trust that the app isn’t stuck
- Trust that what you see is the real final answer
- Trust that restarting won’t break your flow
Thank you
Huge thanks to everyone testing edge cases and sharing detailed feedback.
If you notice anything odd in v0.11.0, please post it — your reports directly shape the next release.
1 Like
Quick update: Informity AI v0.12.0 is out.
This release is mostly about polish and usability:
- Cleaner, more consistent visual themes (including updated naming and better defaults).
- A new Sage theme.
- Improved readability in chat responses (especially for lists and math/formulas).
- Better handling of formatting edge cases so answers look more natural.
- Tuning to make assistant responses feel more direct and helpful.
No major workflow changes in this one, but overall the app should feel smoother and easier to use day to day.
Version: v0.12.0
2 Likes
It would also be nice for the app to prompt for an update whenever available.
Also, can you add the link to the website in your profile here? It’s easier for anyone here to simply click on your profile picture and go to the link.
2 Likes
Maybe I’m missing the point here but why not just read documents or use traditional fuzzy search for documents, like is already built into Finder (since this is a Mac app)?
1 Like
Because Finder doesn’t work in the capacity of a generative AI for your questions about your documents.
2 Likes
Thanks — website link added to my profile!
Update prompts are on the roadmap. Given Informity’s privacy-first stance, the implementation will be user-initiated (a “Check for Updates” button in Settings and/or file menu) rather than automatic background checks. It fetches only a public version file — no telemetry, no data sent. An opt-in automatic check may come later for those who want the convenience. Appreciate the nudge!
2 Likes
Fair question. Spotlight and Finder search find documents containing keywords, they’re good at that. What Informity AI does differently is let you ask questions in plain language and get a synthesized answer pulled from across your entire document library, with citations back to the exact source.
“What did the contract say about termination clauses?” or “which of these papers mentions adverse effects?” - that’s not a search problem, it’s a retrieval and synthesis problem. Traditional search returns a list of files. Informity returns an answer.
Also worth noting that Mac is just the starting point. Windows and Linux versions are on the roadmap, so this isn’t meant to be a Mac-native alternative to Spotlight specifically.
3 Likes
I would like to be able to quickly delete and download another model easily. Right now, I’m not even sure how to delete the model downloaded to free up some space should one need to.
Please see that it’s easy to do it in the future in the app. For now, how can I delete the downloaded AI model?
2 Likes