# Brainfreeze: Building a Source-Traceable LLM-Driven Knowledge Graph System

> Brainfreeze is an Obsidian plugin that implements an enhanced LLM Wiki mode. Through source tracing, reasoning DAG drift detection, and health scoring mechanisms, it helps users build verifiable and auditable knowledge bases locally.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-04-14T08:17:06.000Z
- 最近活动: 2026-04-14T08:25:07.632Z
- 热度: 159.9
- 关键词: 知识管理, Obsidian, LLM, 来源追溯, 知识图谱, 个人知识库, 信息验证, AI辅助
- 页面链接: https://www.zingnex.cn/en/forum/thread/brainfreeze-llm
- Canonical: https://www.zingnex.cn/forum/thread/brainfreeze-llm
- Markdown 来源: floors_fallback

---

## [Introduction] Brainfreeze: Building a Source-Traceable LLM-Driven Knowledge Graph System

Brainfreeze is an Obsidian plugin designed to address the source-forgetting issue of traditional Personal Knowledge Management (PKM) tools and the trust crisis in LLM-assisted knowledge organization. Through source tracing, reasoning DAG drift detection, and health scoring mechanisms, it helps users build verifiable and auditable local knowledge bases, achieving a balance between AI assistance and human judgment.

## Background: The Dual Dilemmas of Knowledge Management

In the era of information explosion, traditional PKM tools (such as Obsidian and Notion) face the problem of source forgetting; while LLMs can automatically extract information, their "black box" nature makes content accuracy and reasoning processes untraceable. Brainfreeze was created to solve these problems, providing a complete methodology for knowledge source tracing and verification.

## Core Mechanisms: Source Tracing, Reasoning DAG, and Health Scoring

1. **Source Tracing**: Label facts with extraction-type ([^e]), reasoning-type ([^i]), and ambiguous-type ([^a]) tags; reasoning-type facts need to mark parent sources to form a chain. 2. **Reasoning DAG and Drift Detection**: Build a reasoning graph, calculate depth (warning if over 2 layers, error if over 3 layers), and detect orphaned reasoning. 3. **Health Scoring**: Combine indicators such as average reasoning depth and proportion of overly deep pages; a score from 0 to 100 reflects the health of the knowledge base and triggers a reconstruction signal.

## Workflow: A Closed Loop from Import to Graph Construction

1. **File Import**: Calculate hashes for deduplication, call LLM to generate drafts with source tags. 2. **Manual Review**: Drafts are stored in .drafts/; the review panel supports source checking and approval of items one by one or in batches. 3. **Graph Construction**: Merge drafts, update links and indexes, recalculate health scores, and form an interconnected knowledge network. All data is stored locally to protect privacy.

## Technical Implementation: Double-Layer Indexing, Structural Checks, and Reconstruction

1. **Double-Layer Search Index**: Structured index (YAML metadata) + full-text index (FlexSearch), balancing precise and fuzzy queries. 2. **Structural Linter**: 12 local checks (broken links, isolated pages, etc.) with zero API cost. 3. **Reconstruction Operation**: After archiving and backing up, re-ingest original files, generate new Wiki pages, and correct systematic biases.

## Application Scenarios: Knowledge Management Solutions for Multiple Domains

Applicable to scenarios such as academic research (paper extraction and citation relationship establishment), technical document management (traceable decisions), personal learning (knowledge point organization and system reconstruction), and news public opinion analysis (event context tracking and contradiction identification).

## Design Values: Human-Machine Collaboration and Local First

Core values include: human-machine collaboration (AI-assisted extraction, human responsibility for judgment), verifiability first (each statement can trace its source), local first (data stored locally), and progressive improvement (health indicators support continuous optimization).

## Conclusion: A Prudent Approach to Knowledge Management in the AI Era

Brainfreeze helps users maintain control and verification capabilities over knowledge while enjoying the efficiency of LLMs. For Obsidian users, it is not just a plugin but also a knowledge management mindset of recording sources, marking reasoning, and conducting regular audits, helping to stay clear-headed in the AI era.