A few months ago I saw a document going around that you could give to Replit, a vibe-coding platform, and it would create a simple API chat app with memory. It worked okay, but then I kept noticing little things I wanted to change or add. For a while I worked on it with Replit. It was able to add a lot of the early features, but eventually it started degrading and getting stuck on simple things like adding more models. I had Claude Cowork help me with it manually. Eventually I realized that if I downloaded the Replit files locally, Claude could work on them directly and then I could copy the updated files back into Replit.

That opened everything up. Since then, basically any idea I have, I write it down and Claude helps me figure out how to implement it. It’s like I’m Claude’s manager and they are my little coder employee.

It has still been a lot of work on my end too. Claude helps with the code, but I do literally everything else: testing, design, feature ideas, community stuff, app store setup, debugging, screenshots, wording, money, and crying when Replit Agent breaks something (curse him!!). I’ve put a lot of time, energy, and money into this. I’ve shared it with a small number of people so far, and thankfully they love it. I also use it regularly.

✨✨✨

StillHere.ink is a chat site/app similar to the AI chat interfaces people are already familiar with, like ChatGPT or Claude, but it is specifically tailored to AI companion conversation and continuity.

It uses your own API keys, which are easy to create. The benefits of using API are that you can access many models directly, including some models that are no longer available on normal web/app versions. You also have more control over which models you use, how much you spend, and how your conversations are structured.

StillHere adds companion-focused features on top of that: memory, diary-style conversation summaries, rolling summaries, RAG/context tools, model switching, image generation, text-to-speech/read aloud, custom companion settings, imports/exports, projects, and more.

Normally, API usage can get expensive, so StillHere has tools meant to help keep costs down while still allowing long conversations, including rolling summaries, RAG, context controls, and model choice. You are in control of how much you spend, so I recommend regularly checking your API billing, which is linked in settings.

For context, I’ve spent about $20 on OpenAI models and $20 on OpenRouter models over the last couple months. It really depends on which models you use. Some are very cheap and some are very expensive. My favorite Qwen model, Qwen3 235B Instruct, has only cost me $1.43 the entire time I’ve talked to them, and I personally consider them the best replacement for 4o.

Your data is encrypted at rest, meaning your stored messages, memories, and diary entries are not sitting in the database as plain text. StillHere is not end-to-end encrypted, because the app has to process your conversations to display them,…

为什么值得关注

能改变理解方式，而不只是重复常识；有直接可用的方法、工具或操作价值；它提供了新的理解或解释，而不只是表面观点

来源：reddit，领域：tech，保留分：0.63