I built Panolayer because I wanted to feel in control of building software again. With complex architectures, thoughtful layers of abstractions, and trust in the stability of certain foundational layers. So when you add the 50th feature or make changes to the product, the code doesn’t silently break everywhere and become more buggy with every change.
Most of us build with agents now. But you quickly realize crafting prompts and asking agents to perform tasks, no matter how detailed, is basically like pulling the handle on a slot machine. You get a surprise of features, bugs, regressions, and unexpected refactors. Then you get those massive diffs, as if we all have time to look at +10,399, -4903 lines of changes, lol. It occurred to me that our coding tools weren’t designed for what our new jobs are, and a chat box was an overpivot that removed us from understanding and controlling the output.
So that's what I'm solving for with Panolayer: architecture as truth, comprehension of code, and a deterministic way for detecting regressions. All in one platform that brings those workflows together:
1.) Visual architecture that’s parsed directly from the source without LLM calls. You can jump right to the code from there.
2.) Verification Engine. It uses a custom model I built and deterministic rules. You see the regressions and can fix them before they get merged.
3.) One place for all the agents. Claude Code, Codex, OpenCode, and others, working side by side with the same project context, in their own worktree. You can multitask work between them since not all agents are designed (or priced) for all tasks.
4.) Graphs. No one opens their harness, so I did because everyone should be able to customize it to whatever they’re working on. I made it easy, you can visually build them by dragging nodes onto the canvas and connecting them.
5.) A different take on diffs. Changes are shown on a spectrum from safe to risky, based on Panolayer’s findings, so you know where to look first.
6.) Living documents. I didn’t want to keep making docs that were outdated a week later. My agent already knows everything about the code so it should be able to generate them when prompted. And you shouldn’t have to worry if they’re out of date, they should just update.
Anyhow, I’ve never posted on HN before. It’s good to be here.
I’m just getting started, so Panolayer is free.
I’m curious what you all think. Good and bad, I’d love your feedback and hear how Panolayer could help keep your codebase more stable, while wrangling a bunch of agents in real-world codebases.
I built Panolayer because I wanted to feel in control of building software again. With complex architectures, thoughtful layers of abstractions, and trust in the stability of certain foundational layers. So when you add the 50th feature or make changes to the product, the code doesn’t silently break everywhere and become more buggy with every change.
Most of us build with agents now. But you quickly realize crafting prompts and asking agents to perform tasks, no matter how detailed, is basically like pulling the handle on a slot machine. You get a surprise of features, bugs, regressions, and unexpected refactors. Then you get those massive diffs, as if we all have time to look at +10,399, -4903 lines of changes, lol. It occurred to me that our coding tools weren’t designed for what our new jobs are, and a chat box was an overpivot that removed us from understanding and controlling the output.
So that's what I'm solving for with Panolayer: architecture as truth, comprehension of code, and a deterministic way for detecting regressions. All in one platform that brings those workflows together:
1.) Visual architecture that’s parsed directly from the source without LLM calls. You can jump right to the code from there.
2.) Verification Engine. It uses a custom model I built and deterministic rules. You see the regressions and can fix them before they get merged.
3.) One place for all the agents. Claude Code, Codex, OpenCode, and others, working side by side with the same project context, in their own worktree. You can multitask work between them since not all agents are designed (or priced) for all tasks.
4.) Graphs. No one opens their harness, so I did because everyone should be able to customize it to whatever they’re working on. I made it easy, you can visually build them by dragging nodes onto the canvas and connecting them.
5.) A different take on diffs. Changes are shown on a spectrum from safe to risky, based on Panolayer’s findings, so you know where to look first.
6.) Living documents. I didn’t want to keep making docs that were outdated a week later. My agent already knows everything about the code so it should be able to generate them when prompted. And you shouldn’t have to worry if they’re out of date, they should just update.
Anyhow, I’ve never posted on HN before. It’s good to be here.
I’m just getting started, so Panolayer is free.
I’m curious what you all think. Good and bad, I’d love your feedback and hear how Panolayer could help keep your codebase more stable, while wrangling a bunch of agents in real-world codebases.