Show HN: Scryer – Visual architecture modeling for AI agents (github.com)

3 points by prohobo 4mo ago ↗ HN

I've been working on this desktop tool (FSL license, free for commercial use) for the past month because I now spend more time in a terminal prompting Claude Code instead of using a code editor. It generally works quite well if I ask the right questions, but I still often find a lot of dead code, stubs, or poor architectural choices when I finish a session, and understanding the codebase itself can be jarring after making major changes through vibecoding.

The idea for Scryer is to provide a visual surface using C4 model diagrams to share with an AI so that we both understand the actual state of a codebase, and how proposed changes would affect it. It's basically model-driven development (like UML) but adapted for the LLM era. Because of that, I think using opinionated C4 (https://c4model.com/) is the best approach:

- It's simple enough to understand without putting the developer into a coma

- There's just enough context to guide the AI coherently

- Doesn't try to replace code, but defines structural guardrails and scope

Also, I've included some newer agent-oriented methodologies like "always/ask/never" contracts (which have turned out to be very useful), task decomposition, MCP + ACP connections, etc.

This is very experimental and early, so it's quite rough around the edges, but I'm already using it in my own dev workflow and I hope you guys check it out. I honestly think this might be the year of MDE/MDD - as we abstract away the code layer, software architecture fundamentals are becoming more important than ever.

2 comments

[ 3.2 ms ] story [ 17.4 ms ] thread

jtbetz22 4mo ago ↗

The architectural drift problem you're describing is exactly what I see with AI-assisted development — agents move fast but conventions slip. I built Caliper to catch that at the hook level: it reads your CLAUDE.md (or existing .cursor/rules) and runs deterministic convention checks after every Claude Code turn, completely free and sub-second.

For the messier stuff — logic bugs, security issues, spec drift against your project's actual policy — there's an optional AI review layer that evaluates changes before commit. It learns from feedback and runs entirely in your environment. You bring your own API key, and select the model to use, so you control the costs.

The two feel complementary to me: Scryer gives you the visual understanding of what changed architecturally, Caliper enforces that it stays compliant. If you're exploring this workflow, we're actively looking for alpha testers.

https://getcaliper.dev has docs and the install is just `npm install --save-dev @caliperai/caliper && npx caliper init`.

VaiPai15 4mo ago ↗

The visual modeling approach makes a lot of sense for designing agent systems, being able to see the whole architecture before you build it is underrated. Scryer looks like a solid tool for that design phase. The next logical step from visual modeling is having the diagram actually become the system. That's what Lyzr Architect (architect.new) does from the other direction, you describe what you want to build in plain English, and Architect generates both the agent graph and the working full-stack application from that. The PRD and agent architecture are visible at the plan phase, so you can see and approve the structure before a line of code is written. Curious whether Scryer is thinking about execution, or staying as a design/documentation layer?