This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady
What I Built
Unison is a 100% local, air-gapped engine that intercepts SMS, WhatsApp, emails, and system alerts, stripping out the noise to build a single consolidated dashboard. Instead of letting real-time pings hijack your attention across five different apps, it batches the clutter, deduplicates overlapping messages using local vector embeddings, and runs a lightweight LLM (like Qwen2.5-3B via Ollama) to synthesize the chaos into one-sentence action items. This permanently kills the mental tax of context-switching and manual digital triage.
Anyone suffering from notification fatigue gets to be significantly lazier. You no longer have to open apps, scroll through hundred-message group chats, or filter spam just to extract the actual signal. Unison handles the doom-scrolling for you and hands you the bottom line in the background, freeing you to ignore your phone without the anxiety of missing something critical.
Code
🦥 Unison
Slow and Steady Notification Intelligence.
A private, local-first communication aggregator, semantic deduplicator, and AI digest engine. Built for quiet minds and zero-distraction focus.
📸 Overview
Modern digital communication forces us to constantly context-switch across dozens of messaging platforms, email inboxes, and system popups. This creates notification fatigue, fragmented attention, and chronic cognitive overload.
Unison flips this dynamic on its head. Instead of reacting to endless real-time pings:
- Passively Ingests alerts from WhatsApp, Android SMS, emails, and system feeds.
- Semantically Deduplicates floods of incoming chatter using local vector embeddings.
- Triages & Summarizes messages into clean, actionable digests using offline local LLMs.
- Presents a Unified Dashboard accessible seamlessly across both Desktop and Android devices.
✨ Key Features
- 🔒 100% Local & Air-Gapped: Zero telemetry, zero cloud API calls. Notification bodies, credentials, and contacts never leave your device.
- 🧬 Intelligent Vector Deduplication: High-frequency chat floods (e.g. 15 messages in 2…
How I Built It
Unison is built entirely on a local, air-gapped open-source stack to guarantee zero data leakage and eliminate cloud API dependencies. The system relies heavily on local inference and embedded databases to keep resource overhead low.
The core intelligence runs on Qwen2.5-3B-Instruct served locally via Ollama. This handles the actual reasoning and summarization. To prevent the LLM from being overwhelmed by rapid-fire pings, the ingestion pipeline routes raw notifications through FastEmbed (using the BAAI/bge-small-en-v1.5 model) to generate text embeddings. These are stored in LanceDB, an embedded vector database. This setup allows the system to vector-deduplicate overlapping alerts—like ten rapid messages in a single WhatsApp group—into a single context block before passing it to the LLM.
To ensure the LLM output is actually usable by the application, the processing engine uses the Instructor library paired with Pydantic v2. This forces the Qwen model to return strictly validated JSON schemas, classifying the priority of the thread and generating a concise one-sentence action item. The entire processed dataset is then rendered into a unified, single-page dashboard using Streamlit, avoiding the bloat of a complex web framework while keeping the interface functional and clean.
Why Does Open Innovation Matter?
Open innovation is the only reason Unison can exist as a viable product. Notifications contain a user's most sensitive, unfiltered data—from bank OTPs and medical reminders to private WhatsApp chats and confidential work emails. Building this with a closed API would require streaming a user's entire digital life to third-party corporate servers, creating a massive privacy violation and security risk that destroys user trust before the app even opens.
By leveraging open-weight models and open-source frameworks, Unison enables zero-trust, air-gapped processing directly at the edge. It also makes continuous, 24/7 background summarization financially feasible, bypassing the massive API costs associated with parsing thousands of daily pings. Open-source AI allowed us to take an architecture that would normally be a privacy nightmare and turn it into a secure, self-sovereign tool.
The Lazy Payoff
The best way to prove a productivity tool works is to dogfood it, so I used Unison to survive the week while building Unison.
Over the last few days, my devices were bombarded with hundreds of notifications—ranging from Discord pings and GitHub alerts to endless WhatsApp group chatter and promotional emails. Instead of breaking my flow state to check my screen every time it lit up, I let Unison run in the background.
It successfully condensed a full day of digital chaos into 8 to 10 actual, single-sentence action items. I restricted myself to checking the dashboard only twice a day. Quantitatively, it saved me roughly an hour of active reading and context-switching daily. Qualitatively, the ROI was massive: it completely eliminated the anxiety of the "unread badge" and proved that you don't need to be hyper-connected in real-time to stay on top of your responsibilities.
Top comments (0)