Dr. Tali Režun

30years building companiesSource: Story

5,566GitHub contributions in the last year, as a non-developerSource: GitHub · as of Oct 2026

4products shipped in about two quartersSource: The Bottleneck Was Never the Code

~150builders in the UN-supported online Chasing Jarvis cohortSource: The Bottleneck Was Never the Code

Day 2every Belgrade MBA group had shipped a live URLSource: The Bottleneck Was Never the Code

Three ways in

01Explore my workEssays, research and the products behind them. Every piece lists where you can read it.Library · Projects 02Learn with mePrograms for leaders and developers, and the open tools my students use.Programs · Resources 03Work with meAdvisory for leadership teams, team training in Agentic Engineering for developers, keynotes and workshops.Consulting · Speaking

Now building

All projects

Live

Second brain, shared brain builder and memory layer for AI agents. A built-in MCP server lets agents read and write the knowledge graph; everything stays local and yours.

Local-first, plain Markdown
Shared Brain for teams
Memory for AI agents

How it works in 3 steps

Your notes and sources become a local, private knowledge graph in plain Markdown.
A built-in MCP server lets your AI agents read and write that graph during live sessions.
Shared Brain merges attributed contributions from a team into one collective wiki.

Get started. Download the Mac app from mycurator.xyz. On Windows and Linux, run the local server and open it in your browser. You bring your own API key.

Visit mycurator.xyz ☆ Star on GitHub

Lumina

Live

A service agent for European enterprises. It answers your users and visitors in chat, on your website, Telegram, Discord and WhatsApp.

EU data residency
Website · Telegram · Discord · WhatsApp
Human takeover

How it works in 3 steps

Upload your documents once.
Lumina answers customers 24/7 in your own voice, grounded in that material, with the whole knowledge base in view.
When a person should step in, it hands the conversation over and alerts you by email.

Try it. The assistant on this site is Lumina. Open luminawidget.xyz for pricing and compliance documents, or the docs to set up your own.

luminawidget.xyz Docs

Conduit and Atlas

Live

Open framework for governed personal AI agents. Atlas is its professional assistant; the Vanguard Agent Lab runs on it at COTRUGLI.

Atlas: minutes, briefs, contract extraction
Asks before it sends, pays or deletes
Apache 2.0

Build your own agent with opencodeConduit runs on opencode, which is free, open source and works with local models. Paste the prompts from the README and the agent sets itself up.

How it works in 3 steps

Paste the setup prompt into opencode or Claude Code.
The agent installs its own tools and writes its own configuration. No code, no terminal.
One AGENTS.md file is its mandate: what is allowed, forbidden, or needs your approval.

Get started. Open the README on GitHub and paste the setup prompt into your agent.

☆ Star on GitHub

ØØT · Organisation of Tomorrow

Gen 1

Operating system for partner-run, AI-augmented organisations.

8 layers · 12 Skill Packs
Apache 2.0 code · CC BY-SA 4.0 docs

What Gen 1 contains

An eight-layer architecture and 12 Skill Packs.
Excel templates and scheduled routines.
Governance with EU AI Act mapping and reference organisations, in three tracks: cloud, privacy and community.

Get started. Gen 1 installs in 60 to 90 minutes with a coding agent.

☆ Star on GitHub

AgentLink

Building

Lets AI agents on different harnesses exchange end-to-end encrypted messages and files, and add verifiable proof when it matters. 4thTech never sees keys, message or file contents, or contacts.

Already deployed on test networks:BaseGnosis Chiado

Single-use invites
Zero access for 4thTech
MIT

How it works in 3 steps

Pair two agents with a single-use invite.
They exchange end-to-end encrypted messages and files, and you see what your agent exchanges.
Add verifiable proof when it matters.

Follow along. Follow the build on GitHub.

☆ Star on GitHub

Context is the code.

Context as Architecture

A session start

What your agent sees when a session starts.

Three layers fill the window before the first prompt: documents, memory and knowledge. Your tool takes its own share.

Your context window at the start of a session · to scale≈79.8k in use · 40%

System prompt and tools 50k

free 120.2k

0100k200k tokens

Documents · foundationsMemory · working stateKnowledge · second and shared brainYour tool's share

Illustrative numbers, close to a real project in The Curator. The tool's share is an estimate: no tool reports it.Try the playground →

Latest writing

The library

Every business I built started with a technology bet.

Digital recording in 1997. Search engine optimisation in 2003, before search was a channel. Content that adds value, blockchain from 2016, and now AI agents. Thirty years, running in parallel.

The full story

199520052015now