Safe, private AI features for apps. I build local-first software that keeps the user's data private: the user's own AI key, a safety check on every message and reply, encrypted storage and no user text in logs.
Socrates: a conversation with four artificial Greek philosophers (Socrates, Diogenes, Plato, Aristotle) who ask hard questions instead of giving verdicts. A Go engine, plain-JavaScript screens, a desktop window and an Android app. Free beta for Windows, Linux and Android, in English and Greek. Code
Moly: a communication coach that asks the one question that matters before you send a message. Browser extension (Firefox, Edge), desktop and Android app, on a Go backend. Local-first with encrypted storage. Free and open source. Code
Socrates prototype: the earlier Python multi-agent platform (about 57,000 lines, around 110 test files), split into 13+ packages on PyPI such as socratic-nexus (one client for Claude, OpenAI, Google and Ollama) and socratic-agents. Research code, not maintained as a product.
- Safety first: every message is judged and every reply is checked before it is shown.
- Tests that prove themselves: each protection is broken on purpose to confirm a test catches it. An independent code review of the safety path found nine issues, all fixed.
- Privacy by design: your own provider key, encrypted local storage, profile export and deletion.
Fixed-price work: AI-feature audits, safe LLM integration, and taking web apps to Android or desktop (signing, F-Droid, stores). Contact: EfthimiosAngelopoulos@proton.me · LinkedIn · PyPI · hermesoftware.wordpress.com
Go · Python · JavaScript · Claude and OpenAI APIs · SQLite · Android · Wails