You know that feeling when you run /new and realize you’re about to explain your entire project architecture again? For the fifth time this week?
Claude has no idea what you talked about yesterday. It doesn’t know you spent three hours debugging that auth flow, or that you decided to use Postgres instead of SQLite, or that there’s a weird edge case in the payment system you need to work around. Every session starts from zero.
I tried the usual workarounds. I kept markdown files with project context. I wrote summaries at the end of sessions. I built elaborate CLAUDE.md files trying to capture everything Claude might need to know. It worked, sort of. But the overhead was brutal. I spent more time managing context than actually building things. And I still couldn’t find that conversation from two weeks ago where I know I solved this exact problem.
So I built Engram.
The core idea is simple: index every conversation with Claude and use semantic search to pull relevant history back into new sessions. When I start working, Claude automatically gets context from previous conversations - not just keyword matches, but semantically related discussions. That debugging session about auth? It surfaces when I’m working on auth again. The Postgres decision? Claude knows about it without me having to re-explain.
The shift was immediate. I stopped writing defensive documentation. I stopped worrying about context windows. I just… work. When I run /new, I’m not starting over. I’m continuing.
But the bigger change was how Claude started working. With memory, it’s not just an assistant responding to prompts - it’s a collaborator that understands the project history. It makes connections I don’t expect. It remembers constraints I forgot I mentioned. The conversations got better because they’re building on everything that came before.
I feel lost without it now. Working with vanilla Claude Code feels like talking to someone with amnesia. You can still get things done, but you’re constantly re-establishing shared understanding instead of building on it.
Context continuity isn’t a clever feature. It’s what makes AI collaboration actually feel like collaboration.
Engram is open source. If you’re building with Claude and context loss is killing you, give it a try.