Self-host a private AI companion with long term memory that recalls and merges past conversations.
Use the co-reading mode to read a book alongside the AI with highlights and inline discussion.
Run the same companion on the web, as an Android app, and via a Chrome side panel extension.
Swap in any OpenAI compatible LLM provider instead of the default SiliconFlow connection.
| jade3551/sora-mem | adityasharmadotai-hash/docs-reader-rag-agent | ahang008/ah-talent-finder | |
|---|---|---|---|
| Stars | 29 | 29 | 29 |
| Language | Python | Python | Python |
| Setup difficulty | hard | easy | easy |
| Complexity | 4/5 | 2/5 | 2/5 |
| Audience | developer | vibe coder | general |
Figures from each repo's GitHub metadata at analysis time.
Requires your own VPS, domain, SSL certificate, PostgreSQL with pgvector, and an LLM API key, no hosted or free deployment is provided.
Sora-mem is a self-hosted AI companion with long term memory that you run on your own server. The project is fully open source across its backend, web frontend, an Android app shell, and a Chrome side panel extension, but the README is direct that it only provides the code: there is no hosted version and no free deployment, so you need your own VPS, domain name, and SSL certificate to run it. The companion supports streaming chat with the ability to understand images sent in conversation. Its core feature is a long term memory system built on vector search, with older memories fading in relevance over time and similar memories automatically merging together. It also has a co-reading mode for reading a book side by side with the AI, complete with highlighting and inline discussion, an automatic journal that writes daily, weekly, and monthly summaries, and an artifact canvas for versioned HTML, Markdown, or code files that preview inside an isolated sandboxed subdomain for safety. Conversations automatically compress their context window as they get long. The same frontend code is reused to build an Android app through Capacitor, and the Chrome extension lets you read whatever webpage you are viewing directly to the AI with one click. On the technical side, the backend runs on Python 3.11 or newer with FastAPI, PostgreSQL 16 with the pgvector extension for the memory search, SQLAlchemy, and APScheduler for background jobs. The frontend is React 19 with Vite and TypeScript. By default it connects to SiliconFlow, an OpenAI compatible LLM provider, but any OpenAI compatible provider can be swapped in. Setting it up locally involves creating a Python virtual environment, installing backend dependencies, configuring environment variables including a database URL and API key, creating the PostgreSQL database with the pgvector extension enabled, then running the setup script and starting the server, followed by installing and running the separate frontend. Production deployment requires familiarity with nginx, systemd, and PostgreSQL administration, plus setting up a separate sandboxed subdomain for the artifact preview feature. The project is released under the MIT license, and the author notes that some default personality touches reflect their own personal use, which can be changed in settings.
A self-hosted AI companion app with long term vector memory, co-reading mode, an artifact canvas, and Android and Chrome extension clients.
Mainly Python. The stack also includes Python, FastAPI, PostgreSQL.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.