cho.sh
2,238 notes
261021261021Toward Autopoiesis26F6DF오토포이에시스를 향해84A5EE차세대 디지털 브레인 만들기C24273261019261019261018261018디지털 시대의 언어 다양성을 위한 선언문507C2CManifesto for Linguistic Diversity in the Digital Age6B5B52261012261012코리지빌리티1CAB16261005261005261004261004Mesh EngineeringF27D1B그물망 엔지니어링881E3BThe devil is in the details94F308260923260923260919260919260917260917In the Agentic Era, should we become a dumb pipe?D815AC260916260916Coscientist6975BA260915260915Coscientific Swarm5F70E8협력적 연구 군체B9E26A
I prefer TUI

I prefer TUI

Why? Multi-tenant environments. Three layers.

  • End-user UI
  • Agent Runtime Environment
  • LLM Inference Server

Run Claude Code on your MacBook and the first two are always local. The third is usually Claude.ai. SSH into a VPS and install Claude Code there, and the first two move to the server. The third is still Claude.ai. Run Claude RC on a VPS and code from your iPad using the Claude app, and the UI is the iPad, the runtime is the VPS, server still Claude.ai.

Most people physically separate tenancies across laptops, personal and work. In most cases, no big deal.

Need on one setup and it becomes super stressful. Say you have two different toolkits.

  • personal (personal Notion, personal Sentry, personal Linear)
  • workplace (company Notion, company Sentry, company Linear)

Most auth states or code harnesses don't support profiles. You can only be logged into one.

So a natural evolution was to have both.

  • a personal VPS with all personal toolkits set up
  • a workplace VPS with all workspace toolkits set up

Tenancies physically split. That fixes the profile problem, but the client problem stays.

Look at the layers again. MCP and toolkit auth should always live in the Agent Runtime Environment IMHO. A surprising number of harnesses tie it to the LLM server instead (Codex Apps, Claude.ai Plugins) or to the end-user UI (Claude Desktop, Codex Desktop).

Both break on multi-tenancy.

  • Auth on the LLM server, and you cannot reuse LLM accounts across tenants.
  • Auth in the end-user UI, and you cannot use the same app to access multiple tenants.

The only way to reliably isolate different auth information is this. SSH into a VPS, run Claude Code there, never use LLM-server plugins. UI and runtime are the isolated VPS, and the LLM server holds no information on the tenancy.

This way you can provide different toolkits, creating multiple dev environments.

Backlinks2