Ultramemory
<p> Private AI memory for your Mac with no cloud or account </p> <p> <a href="https://www.producthunt.com/products/ultramemory-remembers-everything?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1169667?app_id=339">Link</a> </p>
Last verified:
What is Ultramemory?
Ultramemory is a free, open-source, fully local AI memory application for Mac that transforms your email, Slack, files, screenshots, and other work sources into a private, searchable database on your device. It continuously indexes Gmail, Slack, Google Drive, Calendar, personal files, screenshots (with on-device OCR), audio meeting transcripts, Apple Notes, and browser history into a single SQLite database called ultramemory.db stored in your home folder.
The core recall feature lets you ask questions in plain language like
Ultramemory pricing
Pricing model: Freemium
$0 Free forever—no subscription, no trial, no pro tier. The tool is completely free because local AI means there's no per-query bill to pass on. It is open source (MIT licensed) with no account required. Download, open, and it's yours immediately.
Ultramemory pros
- Free forever with no subscription, trial, or pro tier
- Fully open source under MIT license—every line of code is public
- 100% local: no cloud, no account, no sync, no telemetry
- Works offline without internet—even on a plane
- 10 connectors: Gmail, Slack, Drive, Calendar, Files, Screenshots, Audio, Notes, Browser, and more
- Screenshots include on-device OCR for text recognition
- Every AI answer includes clickable citations to original sources
- Detects and tracks contradictions between sources automatically
- Single SQLite database (ultramemory.db) in your home folder
- Granular control: pause, scope, or exclude per channel, folder, or person
- Export all data or delete everything in one click
- Search works immediately without AI—deterministic full-text and vector retrieval
- Optional Ollama integration with hot-plug model detection
- Supports multiple local AI models: gemma4:26b (32 GB RAM) or qwen3:4b (16 GB RAM)
- Extractive search still works even if AI model is off or missing
- No per-query bills since AI runs locally on your hardware
- macOS 14+ on Apple silicon with 16 GB RAM recommended
- Version v0.1.4 is current and actively maintained
Ultramemory cons
- Mac-only: requires macOS 14+ on Apple silicon
- 16 GB RAM recommended for local AI—may be heavy on older machines
- gemma4:26b model requires 32 GB RAM for best answers
- Beta software at v0.1.4—may have unstabled features
- Browser connector requires you to allow history access manually
- Must install Ollama separately for conversational AI answers
- No cloud sync means no access from other devices
- Audio connector requires meeting transcripts (not raw audio processing)
- No Windows or Linux support
- Local AI setup requires two commands with brew and ollama
Frequently asked questions about Ultramemory
What is Ultramemory?
Ultramemory is a free, open-source, fully local AI memory app for Mac that ingests emails, Slack messages, files, and screenshots into a private, searchable knowledge base. It answers your questions with citations you can check, and your memory never leaves your Mac.
Where is my data stored?
Everything is indexed into a single SQLite database (ultramemory.db) in your home folder on your Mac. All processing happens on-device with no account, no sync, no telemetry, and no server to send anything to.
Does Ultramemory work offline?
Yes. Search works the moment you install with deterministic full-text and vector retrieval—zero AI needed. With a local model through Ollama, answers become conversational and read only the evidence it cites on your hardware, even on a plane without internet.
What connectors does Ultramemory support?
Ultramemory supports 10 connectors: Gmail (mail + attachments), Slack (channels + DMs), Drive (docs + sheets), Calendar (events + invites), Files (folders you choose), Screenshots (OCR'd on device), Audio (meeting transcripts), Notes (Apple Notes), Browser (history you allow), and more into one local database.
How do citations work in Ultramemory?
When you ask a question in plain language, Ultramemory gives a narrated answer where every claim carries a citation. Click one and the original message, file, or screenshot opens beside it with the full evidence trail in order of events. Sources that disagree are detected, tracked, and shown as contradictions.
Is Ultramemory free?
Yes, Ultramemory is free forever at $0 with no subscription, no trial, and no pro tier. A tool that holds your whole working memory shouldn't have a price tag. Local AI means there's no per-query bill to pass on, and every line of code is public under MIT license.
How do I set up local AI with Ollama?
Setup requires two commands: brew install --cask ollama-app and ollama pull gemma4:26b. Ultramemory detects Ollama automatically and hot-plugs the model. You can choose gemma4:26b (best answers, 32 GB RAM), qwen3:4b (fast and capable, 16 GB RAM), or None (extractive search still works everywhere).
Can I control what Ultramemory indexes?
Yes. Every connector can be paused, scoped, or excluded per channel, per folder, per person. You can pause capture globally, exclude specific people and channels (like #personal DMs with Alex or ~/Banking folders), see exactly what is stored and what never is, and export or delete it all in one click.
What happens if contradictions are detected?
When sources disagree, Ultramemory says so: contradictions are detected, tracked, and shown—not papered over. For example, if a venue contract says 'Booked for 40 attendees' but RSVP sheets show '52 yes responses and counting,' it flags this as unresolved before it becomes a day-of problem and tracks it as an open loop.
What are the system requirements for Ultramemory?
Ultramemory requires macOS 14+ on Apple silicon with 16 GB RAM recommended for local AI. The current version is v0.1.4. It's free, open source, and MIT licensed. For best AI answers with gemma4:26b, 32 GB RAM is recommended; for fast and capable performance with qwen3:4b, 16 GB RAM works.