UBA-BRAIN

A shared markdown knowledge base and rule system that lets any AI CLI on any machine run the same client work with the same rules, memory and skills.

127
steps logged by the agents
Source: UBA-BRAIN work log · 26–29 Sep 2026
39
of 67 drafts approved to final
Source: UBA-BRAIN job folders · 26–29 Sep 2026
3
machines on one shared memory
Source: UBA-BRAIN work log · 26–29 Sep 2026

MarkdownPythonPowerShellBash1Password CLINextcloudModel Context ProtocolClaude CodeCodex CLI

Build something like this All systems

Facts

Status: Internal tool
Built for: A designer serving multiple clients
Inputs: Slack · Telegram
Agents: Claude Code · Codex
Machines: Mac & Windows, synced
Language: Python

FIG. 001 / ARCHITECTURE

One brain, any agent, any machine

Live model

Inputs

Slack request
Telegram voice note
Screenshot

UBA-BRAIN (synced folder)

index.md
client rules
product facts
skills
log.md

Agents · 3 machines

MacBook AirClaude Code
Windows PCClaude + Codex
MacBook ProClaude Code

sync

Outputs

Drafts · 001, 002…
Final (approved)
Log line

request → rules → job folder → drafts → approval → log

job code: CLIENT-CHANNEL-TYPE-YYMMDD

Inputs

Slack request
Telegram voice note
Screenshot

UBA-BRAIN (synced folder)

index.md
client rules
product facts
skills
log.md

Agents · 3 machines

sync

MacBook AirClaude Code
Windows PCClaude + Codex
MacBook ProClaude Code

Outputs

Drafts · 001, 002…
Final (approved)
Log line

request → rules → job folder → drafts → approval → log

job code: CLIENT-CHANNEL-TYPE-YYMMDD

Problem

Running design and automation work across several AI tools (Claude Code, Codex, Kimi Code) and multiple machines meant client facts, brand rules and job-naming conventions had to be re-explained every session, with no shared memory between tools or computers.

System

A single markdown folder (synced via Nextcloud) holds an index, per-client and per-product knowledge files, a shared skills library, and an append-only work log. Every AI tool is pointed at the same root instructions file, so it navigates index → client file → product file before answering, opens or resumes jobs using a consistent naming and folder scheme, and logs every step so any tool on any machine can pick up a task where another left off.

FIG. 002 / HOW IT WORKS

Six steps, every job

01
Request
Slack message, Telegram voice note or screenshot.
02
Context
The client file, category codes, product facts and references are loaded.
03
Job folder
Named by job code, with the original request written down.
04
Drafts
Generated with the client's own account, numbered 001, 002…
05
Approval
The designer approves; the file is copied to final.
06
Log
One line per step: time, tool, machine, client, product, what, path, status.

01
Request
Slack message, Telegram voice note or screenshot.
02
Context
The client file, category codes, product facts and references are loaded.
03
Job folder
Named by job code, with the original request written down.
04
Drafts
Generated with the client's own account, numbered 001, 002…
05
Approval
The designer approves; the file is copied to final.
06
Log
One line per step: time, tool, machine, client, product, what, path, status.

Guardrails

Rules every agent follows

Accounts come only from the client file
One client's credits never pay for another
Over 10 images or any video needs approval first
Keys are read from 1Password at runtime
Shared Drive is read-only
Nothing is deleted without approval

Stack

What it runs on

Claude Code / Codex CLI: agents
Python: job + log scripts
1Password CLI: keys at runtime
Nextcloud: sync between machines

MarkdownPowerShellBashModel Context Protocol

AI Product Visuals Pipeline

Internal tool

A skill-driven pipeline that turns a client's chat request into an on-brand AI product photo, picking the right account, model and reference automatically.

4 brands on one pipelinePythonMagnific REST APINano Banana Pro (Gemini 3 Pro Image)GPT Image 2+3

UBA Panel

Internal tool

A read-only status dashboard that shows whether the UBA-BRAIN system — clients, skills, health checks — is in good shape, from any device on the network.

Next.jsTypeScriptBunTailscale+1