A zero-setup Research engine (RAG from multiple sources + local NLP in pure js),included with A local Deterministic AI synthesis using BM25, Cosine simmilarity etc. Its Experimental and not serious, ITS NOT IN FOCUS ANYMORE!
See the Demo : https://roarx-dev.github.io/Ruminance/
- Retrieval-Based Synthesis: It first fetches text and then answers, fetches maybe purely local or not.
- Symbolic-Statistical Architecture: So its an very simple deterministic AI, there's no LLM under the hood. And everything being done is shown unlike traditional AIs(Not to mention that this doesn't makes this better then LLMs its profoundly useless for real-world tasks)
- Sentence-Level Semantic Granularity: It treats every single sentence as a unique "atom" of meaning. This helps it be incredibly precise so it doesn't get confused by long, rambling paragraphs.
- Multi-Layered Retrieval Engine: It finds what you need in two ways: first, by looking for the exact words you typed, and second, by understanding the general "vibe" or topic, even if the wording is different.And an NLP intent is shown in the debug pannel.
- Maximal Marginal Relevance (MMR) Diversification: This is to kill monotony. It makes sure the engine doesn't just repeat the same idea over and over, forcing it to find fresh perspectives to keep the "conversation" interesting.
- Knowledge Graph (KG) Engine: It builds a family tree of ideas. It links either two objects or subjects with a relation word making a triple.
- Associative String-Chaining: By linking sentences that share similar words, it tries^ to creates a natural flow of ideas that feels like someone reflecting on a topic.(^It's still trying)
- Bridge Detection System: It acts as a matchmaker for ideas. It scans different topics to find hidden common ground, allowing it to jump smoothly between different subjects without losing the plot.
- Grammar-First NLP Engine: It uses basic logic to figure out what's going on in a sentence—like knowing that a word ending in "-ing" is an action to "understand" some basic English.
- Observable Intelligence: Every step it takes, from the way it searches - to the path it takes, is all visible and can be traced back to the source.
- Client-Side Locality: It requires no API keys or setups, no servers and hence no node.js or docker at all.It's cool and tiny try it out.
- /help: To get a list of available commands
- /image [query]: To search for images only
- /search [query]: Force AI retrival from local database only
- /theme [dark|neon|light]: To change themes
- /clear: To clear the chat-interface
- /reset: To perform a hard reset wiping all the locally saved data
- /debug: Toggles debug pannel
- /stats: View current corpus statistic
- /upload: To upload your own data as a knowledge database
- The RAG uses Cloudflare worker backend for scrape.For anyone wanting to use beyond just some expreiment is advised tomake their own worker back end.(which is absolutely free)
- In Chat-Input Advance toggle fetches more data
- Thesaurus button on top-right allows uploading a thesaurus dictionary also available in repo which AI can use for enhanced Vocab
- SAVE BRAIN/LOAD BRAIN helps one save the current session(as a json) locally and continue anytime
- TO START OF JUST TYPE A QUERY IN CHAT BOX,AND SEE IT PULL DATA AND PRESENT TO YOU RAW( AND ALSO IN ITS OWN WORDS WHICH ARE JUST NOT SO STABLE YET)
If you find it interesting please star it and if you wish to use it (or any parts of it) you may do so freely and may aswell mention me if you would like to! :D