Hey everyone,
I've been working on Syne, an open-source AI agent framework where memory is a first-class citizen, not an afterthought.
The problem: Most AI assistants forget everything between sessions. Context window resets, your preferences are gone, you repeat yourself every time.
What Syne does differently:
Unlimited persistent memory — PostgreSQL + pgvector. Semantic search over millions of memories, across sessions, forever.
Anti-hallucination memory — 3-layer filter ensures only user-confirmed facts are stored. No AI assumptions in your memory.
Self-evolving — Ask it to learn a new skill, it writes the code and registers it. No restart needed.
No config files — Everything in the database. Change personality, rules, behavior through conversation.
Multi-model — Switch between Gemini, ChatGPT, Claude, Groq mid-conversation.
$0/month possible — Google OAuth (free) + Ollama local embedding (free) + Docker PostgreSQL + Telegram.
19 core tools — exec, web search, file ops, cron scheduler, voice, sub-agents, and more.
Telegram + CLI — Same agent, same memory, two interfaces.
Quick start:
git clone https://github.com/riyogarta/syne.git
cd syne && python3 -m venv .venv
source .venv/bin/activate
pip install -e . && syne init
syne init handles everything — provider selection, database setup, Telegram bot, systemd service. No manual config.
Links:
• 🌐 Website: https://syne.codes (https://syne.codes/)
• 📦 GitHub: https://github.com/riyogarta/syne
• 📄 License: Apache 2.0
Built with Python, PostgreSQL 16, pgvector. Currently supports 6 LLM drivers. 372 tests passing.
Happy to answer any questions!
Top comments (0)