Open source · MIT · Local-first·Runs on your machine

Your brain. Your team's brain. Your agents' brain.

One wiki of plain markdown files on your own disk. Read into it, chat with it, walk it as a graph, share a domain with your cohort, and give your coding agents a memory that survives the session. Too many moving parts to explain on one page, so ask.

Ask about The Curator

docs user guide shared brain agent memory

Reading the documentation…

You're chatting with an AI assistant · answers come from the-curator documentation · nothing you type is stored on this site

Install How it works GitHub

entity concept summary · plain .md · your disk · no database

02Quick start

Install in the way that fits your machine.

macOS gets a packaged app with a menu bar widget. Windows and Linux run the same app locally in your browser. Any platform can hand the whole install to a coding agent with one prompt.

Recommended

The Mac app

A .dmg for Apple Silicon and Intel, published side by side. Includes the menu bar widget and in-app updates. Pick the build for your Mac on the releases page.

Download for macOS Latest release · Apple Silicon or Intel · you choose on GitHub

One-liner

Browser install with a Dock launcher

Runs the app in your browser and adds a launcher to the Dock. The script is readable before you run it.

curl -fsSL https://raw.githubusercontent.com/talirezun/the-curator/main/install.sh | bash

Read install.sh before running it

Manual setup

Run it locally in your browser

There is no packaged Windows app yet. The Curator runs as a local server on your machine and opens at localhost:3333. Needs Node.js 18+ and git.

git clone https://github.com/talirezun/the-curator.git
cd the-curator && npm install
set CURATOR_NO_OPEN=1 && node src/server.js

Open the Windows setup guide

Easier

Let a coding agent do it

If you use Claude Code, Codex, Cursor or a similar agent, paste one prompt and it verifies Node, clones the repository, installs and starts the server for you.

Manual setup

Run it locally in your browser

There is no packaged Linux app yet. The Curator runs as a local server on your machine and opens at localhost:3333. Needs Node.js 18+ and git.

git clone https://github.com/talirezun/the-curator.git
cd the-curator && npm install
node src/server.js

Open the Linux setup guide

Easier

Let a coding agent do it

If you use Claude Code, Codex, Cursor or a similar agent, paste one prompt and it verifies Node, clones the repository, installs and starts the server for you.

install-prompt.txt · paste into Claude Code, Codex, Cursor or any coding agent

{{ agentPrompt }}

Works on every platform

One prompt, and the agent installs it

The prompt checks for Node.js 18+, clones the repository into your home directory, installs dependencies and, on a Mac, builds the app. It never asks for your API key: you paste that into Settings yourself.

The best route for Windows and Linux, where the app runs in your browser.

Section 20 of the user guide covers this route

Needs one API key: Gemini, Anthropic or OpenRouter Keys stay in .curator-config.json on your machine Cost and free tier

03One substrate, three brains

The same markdown files serve you, your team and your agents.

Nothing is exported, converted or duplicated between the three. A domain you read into today is a domain your cohort queries next week and your coding agent reads next month.

Your brain

Read once. Keep the structure.

Ingest a PDF, a Markdown file or a text file and get a summary page plus the entity and concept pages the source earns, all interlinked. Every new source updates the pages that already exist, so the graph gets denser, not just bigger.

Your team's brain

A cohort that reads together.

Opt one domain into a Shared Brain. Contributors push synthesised summaries to a private repository; the collective wiki pulls back as a read-only mirror. Attribution and GDPR erasure built in.

Your agents' brain

Handoffs that survive the session.

Done, decided, blocked: written as a supersedable handoff scoped per project. A new session, a new model or a new machine picks up where the last one stopped.

04How it works

Eight steps from a source file to a memory your agents share.

05The app

Same app on the Mac and in the browser. Every screen writes files you can open.

Seven tabs on the left rail: Chat, Ingest, Domains, Shared, Memory, Sync, Settings. Three of them below; switch with the tabs in the title bar.

The Curator

06The menu bar widget · Mac app

What your agents are working on, one click from the menu bar.

Every handoff an agent saves through MCP shows up here: which project, which scope, which tool wrote it and how long ago. The save pulse tells you the discipline is holding. Open a scope and the next session, in any harness, starts from that state instead of from zero.

Working onprojects / curator

Save pulse{{ gfxPulse }}7 days · 34 saves

Written byClaude Code · 2 tools

User guide §6b: the menu bar icon →

{{ gfxPulseTiny }} Mon 14 Sep 11:12

Working on: projects / curator4 days ago

Claude Code · fable-5

Save pulse

{{ gfxPulse }}7 days · at least 34 saves · 2 tools

projects / curator · 4 days ago…

{{ sc.name }}· {{ sc.when }}{{ sc.sub }}

Curation, not retrieval. Files, not a vendor.

Everything The Curator writes is markdown with YAML frontmatter in a folder you choose. Open it in Obsidian, grep it, commit it, feed it to a different model next year. Nothing here is locked to one AI, one tool or one company.

Your brain · your team's brain

Knowledge accumulates.

Every source adds to existing pages instead of duplicating them. RAG re-derives knowledge on every query and forgets; The Curator compiles it once, during ingest, and keeps it current, so the graph gets denser rather than just bigger.

Your agents' brain

State supersedes.

Each save replaces the previous handoff, because a resolved blocker must not come back. One supersedable page per project and scope: what is done, what was decided, what is blocked right now.

Retrieval at query time

Scans raw files on every question
Structure inferred, then discarded
Contradictions stay invisible

Curation at write time

Reads once, writes persistent pages
New sources update existing pages
Conflicts flagged where they conflict

domains/articles/concepts/agent-memory.mdupdated 2026-09-10

---
type: concept
title: Agent Memory
domain: articles
sources: 4
tags: [type/concept, agents, memory]
---
Agent memory is the persisted working state a coding
agent carries between sessions: what is done, what was
decided, what is blocked.

Related: [[agent-memory-compounding]] ·
[[session-state]] · [[project-rollup]]
Entities: [[claude-code]] · [[opencode]] · [[codex]]
Summary: [[the-agent-memory-problem]]

Read the essay on why knowledge is kept as files you own →

08Compounding knowledge · articles and research

Every article builds on the last. So does the research.

The thinking behind The Curator was written in public, one essay at a time. Read them in order and the argument compounds the way a wiki does: each one links back to what came before.

All essays on Substack →

research/ · open to contributions

The research lives in the repository.

Articles, notes and experiments behind the design decisions, kept as markdown next to the code. Read it, cite it, or open a pull request with your own findings.

Open research/ on GitHub →

Knowledge immortality and the second brainresearch/articles · the founding essay Working statedocs · handoffs across session, model, tool and machine Shared Brain architecturedocs · concept, decisions, compliance, roadmap Field Notesthe published surface of a Curator knowledge graph

09Who is behind The Curator

Built by Dr. Tali Režun with the help of community.

Dr. Tali Režun lectures at COTRUGLI Business School and built The Curator in cooperation with Co-Lab, the school's innovation lab. The COTRUGLI cohort used it as it was built and shaped what it became.

Early adopters, first users and community members gave the feedback that made it usable. Thank you for the time you invested; it was the most valuable input this project has had.

The project is MIT-licensed, apart from ten source-available Shared Brain backend files. There is no hosted tier and nothing to sell: your API key talks to the provider you chose and nothing passes through anyone else.

Field Notes COTRUGLI Business School MIT licence

·Get in touch

DMs are open. Bugs go to GitHub.

X@talirezun LinkedInin/talirezun Substacktalirezun.substack.com GitHub issuesversion · platform · what you did

Open source. MIT. Yours to fork.

If The Curator is useful to you, a star on GitHub is how other people find it.

Star on GitHub Install

Ask anything about The Curator. The assistant answers from the project's documentation.

Reading the documentation…

Install in the way that fits your machine.

The Mac app

Browser install with a Dock launcher

Run it locally in your browser

Let a coding agent do it

Run it locally in your browser

Let a coding agent do it

One prompt, and the agent installs it

The same markdown files serve you, your team and your agents.

Read once. Keep the structure.

A cohort that reads together.

Handoffs that survive the session.

Eight steps from a source file to a memory your agents share.

{{ s.title }}

Same app on the Mac and in the browser. Every screen writes files you can open.

What your agents are working on, one click from the menu bar.

Curation, not retrieval. Files, not a vendor.

Knowledge accumulates.

State supersedes.

Every article builds on the last. So does the research.

The research lives in the repository.

Built by Dr. Tali Režun with the help of community.

DMs are open. Bugs go to GitHub.

Open source. MIT. Yours to fork.