If you work with multiple AI tools in parallel — ChatGPT for some things, Claude or Claude Code for others — you probably know the problem: every time you switch, you re-explain the project from scratch. That's the gap I'm trying to close with PE_EDITOR, a file-based "working memory" layer that sits in front of whichever AI you're using — no database, no cloud, just PHP + JS + JSON on your own machine.
Before going further, I wanted to check what's already out there. Three categories overlap, but none hit the actual combination I'm after.
Prompt-manager browser extensions (PhraseVault, XXL Prompt Manager, Prompt Genie, PromptPilot) come closest to PE_EDITOR's variables + clipboard layer. They save prompt templates with placeholders and inject them into ChatGPT, Claude, Gemini and others via the clipboard or directly into the composer. PhraseVault even supports dynamic placeholders — {{clipboard}} grabs whatever you last copied, {{input:Label}} opens an input field before pasting — conceptually close to PE_EDITOR's {{PROJECT}} variables. What's missing: flat placeholders only, no composable skill blocks, no visible final document to review before sending, no separation between durable project profile and current working memory.
AI memory layers (Mem0, Supermemory, and newer entries like "Second Brain for AI," explicitly pitched as persistent memory for Claude/ChatGPT/Cursor) are chasing the same "Git is code memory, working memory is missing" problem. The catch: almost all run on vector databases behind the scenes, so from the user's side it's a black box — you don't actually see what gets stored or fed back in. That's the opposite of what I want: visible, editable context you can reason about before it goes anywhere.
LLMOps / prompt-management platforms for engineering teams (Langfuse, Humanloop, LangSmith, Pezzo) are a different audience entirely — API-centric tooling for teams building their own LLM apps, not for someone using ChatGPT or Claude through a browser tab. Worth ruling out explicitly so nobody expects PE_EDITOR to compete there.
What I haven't found anywhere is all four properties together: fully file-based/self-hosted with no cloud requirement, an explicit visible build step you review before sending, composable skill blocks instead of flat variables, and a focus on hybrid/legacy stacks (Harbour/FiveWin + PHP/web, in my case). The ingredients exist scattered across different tools, but nobody's combining them for this audience.
So I don't think PE_EDITOR competes head-on with any of these — it sits in the gap between "prompt browser extension" and "AI memory SaaS," for a niche that none of them are really built for.
Curious if anyone here knows a tool that already does this and I just haven't found it.