Agentic Knowledge Management
Workshops & AI-assisted practices
· ○ ✦ * ·
· * ·
* · ✦
✦ *
· * ·
· ·
╔═╗╔═╗╦ ╦╔═╗╔╗╔╔═╗╔═╗
* ╚═╗╠═╣║ ║║╣ ║║║║ ║╣ ·
╚═╝╩ ╩╩═╝╩╚═╝╝╚╝╚═╝╚═╝
▲
✦ │ *
rise what │ matters
│
· * │ · ·
✦ · │ ◐ *
* │ * ·
· ✦ │ · ✦
Build your Personal OS. AI agents that know your context.
Your data lakes. Your knowledge. Your decisions.
Salience is a hands-on lab for building your Personal OS — an integrated AI infrastructure that unifies your knowledge, health data, communications, and autonomous agents into one queryable, actionable system.
Build a Telegram agent with Claude Agent SDK. Create data lakes from your browser history, health metrics, and meeting transcripts. Let AI surface what matters from 10+ years of notes.
⏺ Extracting transcript, identifying key concepts...
⏺ Creating atomic notes with backlinks to related ideas
✓ Added 12 notes, linked to 34 existing concepts in your vault
⏺ Processing 5 recordings, transcribing and analyzing...
⏺ Write(weekly-insights.md) with action items and decisions
⏺ Analyzing 847 notes in your vault...
Found 23 connection clusters
Generative models + artistic process (8 notes)
Human-AI collaboration patterns (6 notes)
Creative constraints + emergence (5 notes)
⏺ Generated MOC (Map of Content) with visual graph
⏺ Gathering relevant notes (142 found)...
⏺ Cross-referencing with recent papers in your library
⏺ Write(agent-architectures-brief.md)
✓ 3,500 word brief with citations to your sources
⏺ Searching notes from November 2024...
Key insights captured
Working memory limits (Miller's 7±2)
Spaced repetition algorithms
External memory prosthetics
→ Linked to your project on PKM automation
⏺ See: [[memory-systems]], [[spaced-repetition]]
Knowledge workers, researchers, PhD students, startup founders, and lifelong learners. People ready to invest time and attention into mastering the future of knowledge work.
Most participants are not developers. Basic computer literacy is all you need (if you can use a text editor and a browser, you're set).
This lab is for those who value rational, evidence-based approaches. We follow research and best practices from LLM developers (Anthropic, OpenAI) and proven approaches to working with agents. Yes, this is actively researched and published in academic papers.
I recommend a practical approach through your own project. For some, the project might be a Telegram agent, a voice interface, or a full Personal OS. Others might focus on research: technology stack, use cases, constraints and resources.
Real case studies and everyday problem-solving examples will help you understand how AI agents can be useful in your work, personal life, and interests.
All sessions are practice-based. Presentations (made in Claude Code!) are accompanied by live demos. Participants work in Claude Code alongside us.
Live sessions expect active participation. Technical demos are published as short videos between sessions. During live meetings you can use agents in real-time, create prototypes, write and run code — but deep work requires dedicated time outside sessions.
"I've worked in IT for 20 years, building things with my hands. I'm a multipotentialite — one profession is never enough, I have several, and new combinations keep emerging."
AI agents let us leverage all existing technologies — and create new ones on the fly. Since ChatGPT launched, I've been building my own tools: code, automations, experimental interfaces. I share what I learn on Tool Building Ape (Telegram) and GitHub.
I've been running AI labs since 2022. ~350 people have gone through my knowledge management and AI agent workshops.
Previously: CPO at Ozon.travel, Chief of Content at Ostrovok, Tsentsiper, and others.
Hard skills:
Soft skills:
We don't just study technology — we ask big questions about the future. What will the internet look like when millions of autonomous agents are constantly mining it for opportunities and vulnerabilities?
What does it mean to automate 80% of office work?
What remains uniquely human? What's the role of the professional and the generalist in the future?
Claude Code runs in the command line. We'll cover terminal basics: Mac/Windows Terminal, iTerm, Warp.
One of the most powerful AI agents. The main character of this lab.
Build custom agents programmatically. Create Telegram bots, voice interfaces, autonomous systems.
OpenAI's coding agent, Claude Code competitor. Works with GPT-5-Codex and as a cloud service.
Open-source terminal AI agent. Alternative to Claude Code with multi-model support.
Local-first markdown editor. The knowledge layer of your Personal OS.
Build mobile interfaces to your Personal OS. Capture, query, automate on the go.
Model Context Protocol for extending Claude with external tools and data sources.
Chrome history, Apple Health, transcripts, photos — your personal data made queryable.
Hands-free access to your knowledge base. Voice Claude, ElevenLabs, Whisper.
The tool list is endless and specific: new tools emerge constantly
We don't just study tools — they'll be outdated soon. To live in a world with agents, hire them and get results, we need to learn to think systematically.
We'll learn to formulate tasks so technology works for you. Describe tasks, detail context, formulate requirements, think in structures and algorithms. A specification written in natural language — that's what turns an idea into a real product.
Starting February 2026
online
Zoom sessions, Telegram coordination
8 weeks — enough time to build real fluency with AI agents, understand their capabilities and limits, and automate meaningful work with Claude Code and Claude Skills.
Each week: case studies, hands-on practice, Q&A, and discussion of AI tools for learning and focus.
Wednesdays 18:00 CET. Theory + practice sessions. Case studies and implementation with Claude Code.
Saturdays 12:00 CET. 2-hour live coding sessions. Solving real problems with Claude Code.
Week 1
Introduction to Agents
Week 2
Prompting & Context
Week 3
Knowledge Layer
Week 4
Data Lakes
Week 5
Skills & MCP
Week 6
Agent SDK
Week 7
Personal OS Assembly
Week 8
Demo Day
Self-paced with community support
Full immersion with mentorship
For teams up to 7 people
Payment options: Stripe (links above), PayPal, IBAN bank transfer (German account), USDT and other cryptocurrencies.
For companies: we issue invoices from a European sole proprietor (selbständig) with IBAN payment.
Questions about payment? Contact Gleb on Telegram.
Yes, we issue invoices from a European sole proprietor (selbständig) with IBAN payment. Contact Gleb on Telegram with your company details.
Claude Code supports Windows, macOS, and Linux. You'll need admin access to install software. If using a work computer, confirm installation is permitted.
The lab welcomes both developers and non-coders. We cover terminal basics and show no-code alternatives (Claude artifacts, NotebookLM, GPT Agents). We expect basic computer literacy - web browsing, office software, and familiarity with ChatGPT.
Live online sessions with screen sharing. All materials and recordings available in your personal workspace. Theory and practice are separated - short assignments follow each session.
You'll need a paid Claude subscription (Pro, Max, or Team). API credits for Claude Agent SDK are separate.
We recommend 3-5 hours per week for materials and practical assignments. However, you set your own pace.
Yes, the Individual tier includes 4 personal sessions to discuss your specific projects and challenges.
Modern AI agent tools including Claude Code, Claude Agent SDK, MCP servers, opencode, Windsurf, and other current technologies for building Personal OS.
Yes, the program includes practical assignments to build your own agents and solutions using the technologies covered.
Yes, all participants get lifetime access - full markdown archive with transcripts, tool collections, and assignments. Video recordings hosted on YouTube (unlisted links) with no expiration.
Full refund available within the first week, no questions asked.
Key terms and concepts covered in the lab
Agents
AI Agent — LLM with tool access, running in a loop
Subagent — agent spawned by another for parallel work
Orchestrator — main agent coordinating subagents
Task Tool — tool for spawning up to 10 parallel subagents
Agent SDK — library for building production agents
Models
LLM — Large Language Model
Haiku — fast, cheap model for simple tasks
Sonnet — main model for daily work
Opus — top model for complex tasks
Context & Prompting
Token — text unit (~4 chars), basic LLM element
Context Window — agent's working memory (up to 200K tokens)
Context Rot — quality degradation as context fills
Lost in the Middle — info loss in middle of context
Prompt — instruction for the model
Meta-prompting — LLM creates and improves prompts
Software 3.0 — programming via natural language
Claude Code Commands
/compact — compress context
/clear — clear context
/resume — continue session
/plan — planning mode without code execution
/agents — manage subagents
/memory — view and edit memory
/cost — current session cost
AI Capabilities
Extended Thinking — deep reasoning mode
Streaming — real-time output
Checkpointing — automatic file change tracking
Session Management — state persistence between sessions
Tools
MCP — Model Context Protocol, service integration standard
Skills — modular capabilities in .claude/skills/
Hooks — automatic actions on events
CLAUDE.md — project long-term memory file
YOLO-mode — mode without confirmations
Ecosystem
Plugins — extensions with skills, commands, MCP servers
Marketplace — plugin and extension catalog
Slash Commands — custom /commands
Development
Vibe Coding — creating code through dialogue
CLI — Command Line Interface, terminal
Deploy — publish to cloud
Vercel/Netlify — deployment platforms
