Video summary

Мы строили второй мозг неправильно

Main summary

Key takeaways

Technology

Summary

Владилен Минин presents a second-generation “Second Brain”: a personal knowledge system that automatically gathers information from multiple tools and makes it available to AI agents as persistent context.

How It Works

  • A background process synchronizes about 2,000 records across seven sources, including Claude Code, Codex, GitHub, Gmail, Yandex Mail, Notion, and one source whose name is unclear in the subtitles.
  • Connectors collect new information roughly every 15 minutes. The system masks secrets, stores full text and a search index in a local SQLite database, and creates a compact card for each item.
  • Each card includes a title, summary, search questions, keywords, and key details. A low-cost language model prepares the cards; the speaker says the cumulative processing cost was 138 rubles.
  • SQLite’s FTS5 full-text search supports a BM25-based ranking system. BM25 scores keyword matches based on factors such as word rarity, frequency, and document length. Results are also adjusted for freshness: a record’s weight halves every 120 days.
  • The speaker reports that the system returns the top eight cards in about 0.1 seconds, without using an LLM during the search itself. The database is local by default but could be moved to a remote server.

Why This Approach

The speaker compares three ways to give AI persistent memory:

  1. Put everything in the prompt: Uses up context, costs more, and requires manual curation.
  2. Use a vector database and RAG: Requires chunking, paid embeddings, extra infrastructure, and reprocessing when changing models.
  3. Use BM25 keyword search: Avoids a separate vector database and embedding-based retrieval while still ranking relevant records.

He argues that the third option is faster, cheaper, and less dependent on any particular AI model.

Demonstrations and Practical Guidance

  • Automatic knowledge graph: A dashboard displays records, their connections, and their originating sources.
  • Model and account portability: After his Anthropic account was blocked, the speaker connected a new Claude account to the local MCP service. It could retrieve his existing context immediately.
  • Task dashboard: An AI agent gathered tasks from email, GitHub, and AI conversations, then created a dashboard using the speaker’s design system.
  • Implementation walkthrough: The video explains the data flow—connectors, secret masking, card creation, SQLite/FTS5 storage, BM25 search, freshness weighting, and MCP access—and shows configurable model providers and manual synchronization.

The main benefit, according to the speaker, is retaining usable personal context across tools and accounts without having to maintain a knowledge base manually.

Main speaker/source: Владилен Минин, presenting and demonstrating his own Second Brain system.

Rate this summary

Your feedback will help improve summaries.

Improve this summary

Reprocess with a stronger model when the summary feels incomplete or inaccurate.

Pro

Translate summary in another language

Pro

Ask questions to this video

Chat for follow-up questions, clarifications, and source-backed answers.

Coming soon

Share this summary

Original video