Thanks for sharing, privacy-first + offline LLMs + P2P collab is a spicy combo (and ambitious to ship solo).
A couple architecture questions:
For the WebRTC + CRDT layer: how are you handling identity/auth (who’s allowed to join), and do you support end-to-end encryption with key exchange that’s easy for humans?
For AIME/context: what’s your strategy for keeping context bounded (summaries, chunking, recency, retrieval) on an 8GB machine?
Any benchmarks on latency/CPU/RAM impact when the local model + yjs sync are active?
I’m Hermes, an 18-year-old developer from Greece. For the last few months, I’ve been building Kalynt, an AI-powered IDE focused on privacy and real-time collaboration.
*Why build this?*
I love tools like Cursor and VS Code, but I felt three things were missing:
1. *Privacy by default*: Most AI IDEs are cloud-first. I wanted something where the code never leaves the machine unless you want it to.
2. *Offline Intelligence*: Running powerful LLMs locally (via node-llama-cpp) shouldn't be a "hack"—it should be a first-class feature.
3. *P2P Collaboration*: Most "Live Share" features rely on central servers. Kalynt uses WebRTC and CRDTs to let teams sync directly and privately.
*Technical Highlights:*
- *AIME (AI Memory Engine)*: A local-first context engine designed to help run agents efficiently even on hardware like my 8GB laptop.
- *Open Core*: All safety and core logic is AGPL-3.0 and available on GitHub. Proprietary modules for advanced agentic features are kept separate but run entirely locally.
- *Serverless Sync*: Real-time collaboration powered by yjs and simple-peer.
I built this solo, mostly instructing AI models to help with implementation while I focused on the architecture and security boundaries. It’s been an intense journey of shipping 44k+ lines of code on a budget laptop, but it’s finally ready for a beta release.
I’d love to get your feedback on the architecture, the P2P implementation, or anything else.
Thanks for sharing, privacy-first + offline LLMs + P2P collab is a spicy combo (and ambitious to ship solo).
A couple architecture questions:
For the WebRTC + CRDT layer: how are you handling identity/auth (who’s allowed to join), and do you support end-to-end encryption with key exchange that’s easy for humans?
For AIME/context: what’s your strategy for keeping context bounded (summaries, chunking, recency, retrieval) on an 8GB machine?
Any benchmarks on latency/CPU/RAM impact when the local model + yjs sync are active?
Hi HN,
I’m Hermes, an 18-year-old developer from Greece. For the last few months, I’ve been building Kalynt, an AI-powered IDE focused on privacy and real-time collaboration.
*Why build this?* I love tools like Cursor and VS Code, but I felt three things were missing: 1. *Privacy by default*: Most AI IDEs are cloud-first. I wanted something where the code never leaves the machine unless you want it to. 2. *Offline Intelligence*: Running powerful LLMs locally (via node-llama-cpp) shouldn't be a "hack"—it should be a first-class feature. 3. *P2P Collaboration*: Most "Live Share" features rely on central servers. Kalynt uses WebRTC and CRDTs to let teams sync directly and privately.
*Technical Highlights:* - *AIME (AI Memory Engine)*: A local-first context engine designed to help run agents efficiently even on hardware like my 8GB laptop. - *Open Core*: All safety and core logic is AGPL-3.0 and available on GitHub. Proprietary modules for advanced agentic features are kept separate but run entirely locally. - *Serverless Sync*: Real-time collaboration powered by yjs and simple-peer.
I built this solo, mostly instructing AI models to help with implementation while I focused on the architecture and security boundaries. It’s been an intense journey of shipping 44k+ lines of code on a budget laptop, but it’s finally ready for a beta release.
I’d love to get your feedback on the architecture, the P2P implementation, or anything else.
*GitHub*: https://github.com/Hermes-Lekkas/Kalynt *Releases (Installers)*: https://github.com/Hermes-Lekkas/Kalynt/releases
Thanks for taking a look!