Building Knemora: An Open-source, Local-first AI (SI) Second Brain

Description and use

Knemora turns documents and datasets into connected knowledge, enabling visual exploration and querying without cloud inference.

Problem

Need to convert documents and datasets into connected knowledge, explore relationships visually, and query without sending data to cloud inference services.

Outcome

Knemora enables local, privacy‑preserving knowledge management and inference, converting raw documents into a searchable, graph‑based second brain.

Workflow

• 1. Import documents and datasets into Knemora.
• 2. Knemora parses and stores content in Markdown pages with frontmatter and explicit links.
• 3. The system builds a graph view of connections using Sigma.js/Graphology.
• 4. Local inference is performed with llama.cpp and Qwen3‑8B Q3_K_M.
• 5. Retrieval uses BM25 search and bounded graph navigation without a vector database.
• 6. Structured calculations are performed with Polars, keeping results grounded in computation.
• 7. Users query the knowledge base via search, graph navigation, or chat, receiving source‑referenced answers.

Evidence

I bring your documents and datasets; Knemora compiles knowledge into a portable Markdown wiki, connects pages, and makes that knowledge available through search, a native graph, and chat with source citations.

Source

Local inference: llama.cpp with Qwen3-8B Q3_K_M as the default model.

Source

Retrieval: BM25 search and bounded graph navigation, without a vector database.

Source

The stack includes Vue 3, FastAPI, Google ADK, SQLite, and Sigma.js/Graphology.

Source