Your AI chats shouldn’t start from zero every time
AI tools are getting smarter, but memory still feels fragmented.
Every new chat means re-explaining context, re-uploading files, and rebuilding momentum from scratch — especially when switching between ChatGPT, Claude, and Gemini.
That frustration led us to build Lumi.
Lumi adds a persistent memory layer to AI workflows with:
cross-session memory
Vaults to organize context
document & PDF retrieval like a personal RAG
support for ChatGPT, Claude, and Gemini CLI
Still exploring the best ways to solve long-term AI memory.
llmmemory.ai
How are others handling context persistence today?
Trending on Indie Hackers
One thing I’d test early is giving people a visible “memory diff” after each session: what was added, what was updated, and what was ignored. The anxiety with persistent context is not just whether it remembers enough, it is whether it remembers the wrong things silently. If users can prune or approve memory in the flow, Lumi becomes less magical but much more trustworthy for real work.
this is bigggg. i was going to work on something in this direction too. you really mapped out things so well. been to your site and seen the loom vids too. low-key think you might be underselling this as mostly an ai memory-consistent tool though. in all, big ups: i truly likeeeeee lumi!
The hardest part for me is not storing more context, it is deciding what should become durable memory versus just chat exhaust. I would make the onboarding opinionated: ask for one workflow, one repo/docs source, and one "never make me repeat this again" rule, then show exactly what Lumi remembered after the next session.
This is a real pain point. The biggest friction with AI tools now is not only model quality, it is continuity. Every new chat resets context, and every tool switch makes the user rebuild the same working memory again.
I’d probably position Lumi less as “memory for AI chats” and more as a persistent context layer for serious AI workflows. Vaults, document retrieval, cross-session memory, and support across ChatGPT, Claude, and Gemini CLI make it feel broader than a simple chat add-on.
The one thing I’d pressure-test is the brand/domain frame. Lumi is friendly, but the category you are entering is going to be trust-heavy because people are storing documents, work context, and long-term AI memory. llmmemory.ai explains the function, but it may also make the product feel more like a technical utility than a durable workflow platform.
Xevoa .com would fit that broader direction better as a clean AI workflow and context platform brand. Same product, but with a name that gives more room if this expands into team memory, project context, RAG workspaces, or agent workflows later.
This feels like one of those products where the memory layer is valuable, but the brand has to make people trust it before they put important context inside.
you totally get it. this is on point!
Exactly.
The reason I pushed on the brand frame is because this category will probably be judged on trust before features.
For a notes app or small extension, a friendly name is fine. But when the product becomes a place where users store documents, project context, AI memory, and long-running work, the name has to feel more durable. People need to feel like they are putting important context into a real workflow layer, not just another AI helper.
That is where llmmemory.ai feels clear but narrow. It explains the feature, but it does not fully carry the bigger product if this becomes persistent context infrastructure across tools, teams, agents, and projects.
Xevoa.com stood out because it feels broader and more platform-like without losing the AI workflow direction. It gives you room to grow from “memory for AI chats” into a serious context layer people can trust.
I would pressure-test that now, before more users, docs, and product language get built around the current frame.