Local AI for document search — Informity AI, runs entirely on-device

Good feedback — model management is on the roadmap to make easier within the app. You’re right that it should be obvious.

For now, models are stored in ~/.informity/models/llm/. Open Finder, press Cmd+Shift+G, paste that path, and you can delete any .gguf file directly. The app will prompt you to download a model again next time you select that profile.

Alternatively, if you prefer managing models yourself, Informity AI has Ollama integration — install Ollama from https://ollama.com and pull a model with something like ollama pull qwen3:32b. The app is specifically tuned for Qwen3.6 35B, Qwen3 14B, and Qwen3.5 9B for best results, but you can try other models too — it’ll fall back to a generic profile. Results may vary depending on the model’s instruction following and context handling.

2 Likes

Thank you. Very helpful.

2 Likes

Informity AI v0.12.1 is out

Small but useful update. The app now has a built-in update checker so you’ll know when a new version is available without having to check the website manually.

It’s manual, not automatic — you trigger it yourself from the Informity AI menu or from Settings whenever you want to check. It’ll tell you if there’s an update available, confirm you’re already up to date, or flag if something went wrong.

Download the latest at https://www.informity.ai

@CloakedNetizen

2 Likes

Informity v0.12.2 is out

We just shipped v0.12.2 with a focus on stability, smoother scanning, and desktop quality-of-life improvements.

What’s new

  • Fixed an issue where the backend process could sometimes stay running after quitting the app, which could lead to high memory usage.
  • Improved scan reliability for cloud-synced and special folders (including iCloud-like locations).
  • Added a global shortcut to quickly bring Informity to the front.
  • Improved macOS packaging polish, including a corrected mounted DMG icon.
  • Update-check dialog now includes a direct link to release notes.

Why this release matters

This update is mostly about making Informity feel more dependable day to day: cleaner app shutdown behavior, safer scanning in tricky folder setups, and small UX improvements that reduce friction.

Release notes: Releases · informity/informity-ai · GitHub

If you try v0.12.2, we’d love your feedback.

2 Likes

Any plans for third-party audits?

2 Likes

Definitely open to it once the app is more mature and I’ve worked through more of the roadmap. In the meantime it’s fully open source — the code is all there and network behavior is verifiable by anyone who wants to look.

The repo can be found here GitHub - informity/informity-ai: Informity AI · GitHub

2 Likes

Informity AI v0.13.0 is out

v0.13.0

This release introduces an optional MCP Server (Experimental) so trusted AI clients can query your local Informity library, while keeping a strong privacy-first and read-only default posture.

What’s included

  • Added MCP Server (Experimental) to Settings for easy client integration.
  • Added read-only MCP tools for health, file listing, semantic search, index status, and scan status.
  • Kept MCP scope intentionally read-only in this release.
  • Improved retrieval quality by filtering out local upload noise and reducing duplicate result entries.
  • Simplified setup for external clients with a cleaner standard command flow.
  • Added stronger safety controls and operational guardrails around MCP behavior.
  • Added dedicated MCP logging so interactions are easier to audit and troubleshoot.

Notes

  • Privacy model: MCP is disabled by default and should be enabled only for trusted clients.
  • Scope: this release focuses on read-only MCP access; advanced/extended MCP capabilities remain deferred.
  • QA status: validated through expanded MCP-focused automated coverage and end-to-end client checks during this release cycle.

Tag: v0.13.0

1 Like

Just found this project, is this basically a local and private version of Goggle NotebookLM?
I don’t have a mac so I can’t try it for now.

I found NotebookLM really useful and capable but I can’t use it with personal infos and files so having a local running competitor would be awesome!

I suppose we could create different chats for different projects. Are the files shared with Informity AI available for selected chat or globally?

I can’t wait for the other platforms version coming.

2 Likes

I’m struggling to get Informity to scan and index all my files (all PDFs). I have 18 in the folder I want Informity to access and the most it will index is 12. I’ve tried increasing the max indexable file size to 500mb but everytime I click save it returns to 100mb. I don’t know if this is what’s causing my issues or not (as tbf all the files are less than 100mb) but is there anything else I can try?

2 Likes

Yeah, this is a bug I am noticing too.

2 Likes

Could you add Informity to homebrew? It’s a lot easier and simpler to install that way and to also update. Plus, the added benefit of the dev being the one maintaining it on brew.

2 Likes

Thanks for reporting this, and sorry for the friction. I found and fixed a settings bug where Max Indexable File Size could revert to 100 MB after saving. That fix is included in the latest release (v0.13.1), so changing it (up to 500 MB) should now persist correctly.

The “12 files” behavior is usually not a hard indexing limit. Informity can skip files for specific reasons (for example: previous non-retryable extraction failure, timeout, protected/corrupt PDF, or scanner ignore/permission issues). A quick thing to try is:

  • Run Rescan All Files.
  • If the scan shows errors, click Errors on the Dashboard scan status to open the per-file error list and see exactly which files were skipped and why.

For any future issues, it would be very helpful if you could file a report at Issues · informity/informity-ai · GitHub. That helps me track reports, connect duplicates, and ship fixes faster.

2 Likes

Please see my reply to dada_goose.

2 Likes

I would do it in the future. If I add the same homebrew request on Github now, then my alias will not be “anonymous” and become “compromised”.

2 Likes

Thanks for checking out Informity AI, and great questions.

Short answer: yes, Informity AI is aiming to be a local/private alternative for the “chat with your documents” use case people like in NotebookLM.

A few details based on how Informity AI currently works:

  • Privacy model: Informity AI is local-first. Your indexed files, vectors, chat data, and local model artifacts are stored on your machine. It also has an offline/full-privacy mode.
  • Chats/projects: Yes, you can use separate chats for separate projects.
  • File scope behavior (important):
    • Indexed library files are part of your local corpus and can be queried broadly in Researcher mode.
    • You can also scope retrieval to selected indexed files for a chat turn.
    • Uploaded files (+ attach in chat) are temporary and chat-scoped (not global).
    • Assistant mode is non-retrieval; Researcher mode is retrieval-grounded.

So practically: you can do both “global local library” workflows and “this chat only” workflows depending on whether you use indexed files vs chat uploads/scoping.

And yes, today the app is macOS-first. Multi-platform support is planned and definitely a common request.

2 Likes

Totally understandable. That was a general request but, I respect your privacy of course.

I will add this to roadmap.

2 Likes

Thanks. The 100Mb limit is now fixed. The error I’m getting regularly is…

Docling extraction failed: No module named ‘docling datamodel’ OCR fallback also failed No module named ‘docling datamodel’.

1 Like

Found and fixed the issue.

Please re-download v0.13.1 from https://www.informity.ai and give it another try. No version bump, just grab the latest build of the same version.

Let me know if you’re still seeing the error after that.

2 Likes

@Informity github issue updated. Thanks

2 Likes