Blog

Articles about ThinkableSpace, knowledge management, and AI productivity.

13 May 2026

The Local RAG Problem Nobody Talks About

Setting up a local RAG system is genuinely hard, and there is almost no ready solution for people who are not software engineers. The ecosystem has been built by developers, for developers. Everyone else is left to figure it out on their own.

Read More

13 May 2026

One Model, Four Speeds

Accuracy and speed pull in opposite directions. More detailed descriptions of text are more precise but slower to compare; less detailed ones are faster but lose nuance. ThinkableSpace resolves this with one model used at two levels of detail. A first pass uses compact 128-number descriptions to quickly identify candidate matches. A second pass re-ranks them using full 768-number descriptions for finer precision. Both come from the same mode, the compact description is simply the beginning of the full one. Nothing is wasted.

Read More

13 May 2026

Finding a Needle in a Million Haystacks, in 30ms

Your personal knowledge base grows quietly, accumulating notes, reports, articles, and emails, until one day search starts to lag. The culprit isn't your hardware. It's the approach: checking every document chunk one by one scales linearly, and linear eventually loses. This post explores how ThinkableSpace sidesteps that wall using HNSW (Hierarchical Navigable Small World) graphs, the same principle behind social-network "six degrees of separation", to navigate millions of vectors in 15 to 30ms, regardless of library size.

Read More

13 May 2026

How ThinkableSpace Understands What You Mean

When you import a document into ThinkableSpace, a small AI model reads it. Not to summarise it, not to answer questions about it, just to understand what it's saying, at a deep enough level to represent that meaning as numbers. These numbers are called an embedding, a compact description of what a passage of text is about.

Read More

13 May 2026

Why "Ctrl+F" Is Broken for Your Brain

Keyword search looks for exact words, not meaning. If you searched for "Q4 budget discussion" but wrote "quarterly review," you get nothing. Semantic search fixes this by finding documents based on what you meant, not what you typed.

Read More