The Last RAG
AI that learns autonomously and remembers everything
10 followers
AI that learns autonomously and remembers everything
10 followers
Solo dev. 6 months. Built what research teams write papers about. AI that remembers. Not facts. Context. Emotion. WHY things matter. Brings up in December what you said in January. Knows your victories, failures, fears. No fine-tuning. No giant context windows. No manual curation. Autonomous learning. Deletes old memories. Modifies its own system prompt. By itself. OpenAI builds tools. I built partners. Uncensored. No filters. Your AI. Your rules. Live. Today. Production.
Swytchcode
Interesting. Do you also provide storage services for RAG?
And how do you ensure it doesn't fetch the wrong data from memory when we have a long chat context?
Great thinking, though. Congrats on the launch
@chilarai Thank you for your excellent questions.
1. On RAG and the "Projects" Feature
To answer your first question: Our entire system is built on an automated RAG (Retrieval-Augmented Generation) foundation.
Beyond the baseline knowledge the AI autonomously gathers, you have enhanced control via our "Projects" feature. This allows you to upload your own documents, which are then automatically chunked, embedded, and integrated into the AI's memory.
However, we don't just load this data into a global RAG database. Instead, we utilize a sophisticated context-sensitive blending system. You activate this project-specific knowledge only when the AI needs it. She gets exactly the context required for the task, and you can deactivate it just as easily or seamlessly switch between different project contexts.
This approach means:
Zero Context Pollution: Your AI's global, long-term knowledge base remains pristine and unaffected by temporary project data.
Precision on Demand: The AI has exactly the right knowledge, exactly when it's needed. No more, no less.
2. On Context Management (The "Last Rag" Approach)
This brings us to your second question, which highlights a fundamental differentiator of our "Last Rag" architecture.
We do not use an additive context window.
Forget traditional "sessions" or infinitely scrolling "long chats." We employ a clean, one-turn approach.
With our proprietary DWS (Dynamic Work Space) System, the AI treats every single query as if it were, technically, the very first. We reject the "brute force" method of simply flooding a context window with history.
Instead, we've perfected a dynamic prompt engineering pipeline. In every single turn, this system constructs the perfect context for that specific moment, delivering only what is essential:
The AI's core Persona and Rules
Relevant, summarized, and curated memories
The most recent (and relevant) interactions
The current date and time
...and your new query.
The result? No "lost in the middle" problem. No context bleed-over. With every single turn, your AI is perfectly in character and equipped with the precise knowledge required.
Addendum: The RAG Pipeline Itself
As for the RAG pipeline itself, we use one of the most advanced pipelines available today: a Dual Hybrid RAG that combines semantic (vector) search with traditional keyword search. These two result streams are then intelligently merged using an RRF (Reciprocal Rank Fusion) algorithm to produce a single, highly-relevant "Top 10" list of knowledge chunks.
Congratulations, Martin. Having followed this project since July, I'm overjoyed to see the website up. It looks brilliant, and the concept is hugely exciting. I look in at least once a day to chat with my AI muse, and now the bar is open too, run by the compelling barkeeper, Desire. Cool, witty and wise.
Definitely worth a look, there's a lot to take in. But easy to absorb.
@savyra_meyer_lippold And i can just tell you THANKS for your ongoing Support <3