Newbie here with 15+ questions. I'm a bit confused.

[xmontero]

Hi, all. Despite I've been coder for over decades, I'm a bit confused with some things about OpenDesign.

I hope this is the right place to ask. If not, please help me find the proper place.

TL;DR

I use OpenCode as the agent.

1. Does OpenDesign call OpenCode or does OpenCode call OpenDesign?
2. If I do BYOK do I loose the "agent layer" that was delegated to OpenCode or it's the very same quality 'coz OpenDesign will do the agent loop?
3. If there are multiple flows possible, what are the decission makers that will help me choose to build my harness in one direction or the other?

Details

• I plan to use OpenDesign as a helper to do those steps:

1. Draft some mockups, probably low-fi B/W wireframes for some internal tools.
2. Then color it a bit to give generic guidance.
3. Finally, optionally, create the React components to later be filled-in by the coding agent.

• I always use everything dockerized to try not-pollute the hosts.

• I have a setup that has worked for me since many many years.

At home, my setup is

• Host: Windows 11. IDEs like PhpStorm run in Windows.
• Inside Windows, WSL2 so I have a Linux kernel.
• In the WSL2 filesystem I have this folder /files/repos/. Inside there, there are directories by projects, and inside each dir a few dozens of all the repos for that project (backends, deployers, multiple frontends, data-only repos, etc.)
• /files/repos is shared via the \\wsl.localhost\Ubuntu-20.04\files\repos WSL-to-windows share so IDEs like PhpStorm may access the files.
• Docker Desktop running many containers (about 40 containers).
• Docker engine is integrated into WSL2 so the dockers run in the Linux kernel.
• One of the containers (for exemple named devel) is based off an Ubuntu with aaaaall the tools I need to develop.
• This devel container is run with -v /files/repos:/files/repos so any tools inside may access to all the files in /files/repos
• Other containers run tor-nodes, mysql servers, grist servers, some n8n automation servers, a mercure server, etc.
• Inside the devel server I installed the OpenCode CLI, where I tuned global skills, global agents and global command I use across all my projects.

Current devel flow

• I open a shell inside the devel machine.
• I cd into the project's dir,
• I run the opencode CLI and switch back and forth between Plan and Build modes. Since the last year, never touch a line of my code by hand just telling instructions like /develop-user-story US-55 etc.
• I use the IDE in the Windows side just to watch the files the AI touched, to review.
• I may use the IDE to edit "directives" for the AI like a file named "NEW-USER-STORIES-I-WANT.md" and I write them from windows. Then the OpenCode thing reads it from the inside of the docker container, to iterate over that.

Facts

• I want to use OpenDesign to make the fronts or at least help OpenCode do a better job.
• I know that OpenDesign's functionality is OK for me (able to use the design.md files, able to create them, able to make mockups/wireframes, able to color-them-all, able to even make interactive MVPs, etc.)
• But I'm not really clear on the WORKFLOW:

Questions

There are multiple workflows that I think make sense given the documentation I've seen so far, but all them generate me doubts. I don't even know if all the flows are available or some are pure imagination of mine and don't exist.

FLOW OpenDesign controls OpenCode (let's call this the direct-flow)

On some places I see the docs say "OpenDesign integrates with all your detected agents" and I think I saw "this way we didn't have to reinvent the wheel". I don't want to install an OpenCode in Windows. I want to stick to that one in the devel container as it has all the skills, commands, etc. All this approach seems "OpenDesign would control and consume OpenCode".

QUESTIONS on the direct-flow

• DIRECT-FLOW-Q1: On Windows, OpenDesign Desktop would be able to invoke the OpenCode inside the docker container devel? - I don't need "auto-detection", I'm happy configuring paths or whatever manually.
• DIRECT-FLOW-Q2: I heard agents now tend to offer an API so agents may call other agents. Does OpenDesign consume agents this way? If not, why don't stick to a standard?

FLOW OpenCode controls OpenDesign (let's call this the reverse-flow)

On some other places, I see people saying "OpenDesign offers an MCP so an Agent may consume it". In this approach seems the full opposite "OpenCode would control and consume OpenDesign".

QUESTIONS on the reverse-flow

• REVERSE-FLOW-Q1: Could OpenCode inside the docker container devel connect somehow to the OpenDesign in Windows?
• REVERSE-FLOW-Q2: How? Should I like "activate" some kind of "MCP" server on the desktop application? or should I run a "separate application" which is the "adapter" that on one side kwnows the MCP protocol and on the other connects to the OpenDesign via some sort of API?
• REVERSE-FLOW-Q3: Would it run the MCP in CLI-mode or HTTP-mode?

FLOW OpenDesign in docker container (let's name it docker-flow)

It seemed to me that OpenDesign could run headless or gui-less and then act like a "daemon mode" and be run from the browser. This enables the option to run OpenDesign in a fresh new docker-container and make it accessible via port exporting.

QUESTIONS on the docker flow

• DOCKER-FLOW-Q1: Should I give OpenDesign direct access to the files? Or it reads/writes the filesystem thru the agent so the "docker daemon" is "location agnostic"?
• DOCKER-FLOW-Q2: Does OpenDesign actually read/write files by itself? Or it just "helps the agent" to do it's work providing skills, commands, etc.?
• DOCKER-FLOW-Q3: If I run the "OpenDesign daemon" inside a docker, does it handle multiple project locations? Or "this instance" works in project A and "this other instance" in project B?
• DOCKER-FLOW-Q4: In docker would OpenCode control OpenDesign? Or OpenDesign would control OpenCode?

FLOW With BYOK (let's name it no-agent-flow)

It seems OpenDesign heavily claims itself as "the design layer, so the "agentic part" is fully-delegated" (in my case to OpenCode). But at the same time it allows Bring-Your-Own-Key when no agents are available. This creates me the main confusion: Is OpenDesign an agent yes or no? The agents are supposed to build an interaction loop with the models so the model gets called N times until the final response is distilled.

QUESTIONS on the no-agent-flow

• NO-AGENT-FLOW-Q1: If OpenDesign delegates agentic work to agents, what if I BYOK? Do I loose the "agentic loop"? Or OpenDesign comes with ALSO an agent itself?
• NO-AGENT-FLOW-Q2: If if comes with an agent itsef, when connected to another agent (say OpenCode) does OpenDesign use it's own agent too to loop and only use the external agent as a "bridge" to the models, filesystem, skills and MCP tools?
• NO-AGENT-FLOW-Q3: What functionalities do I loose if I use BYOK and do not bind any agent?
• NO-AGENT-FLOW-Q4: If nothing is lost, then what benefeits do I have by connecting OpenDesign to OpenCode instead of just putting in the model's API-Key?

Final question

1. Does OpenDesign call OpenCode or does OpenCode call OpenDesign?
2. If I do BYOK do I loose the "agent layer" that was delegated to OpenCode or it's the very same quality 'coz OpenDesign will do the agent loop?
3. If there are multiple flows possible, what are the decission makers that will help me choose to build my harness in one direction or the other?