@zaynjarvisOpenViking: an open-source context database for agents (Zayn, ByteDance)
I'm Zayn Jarvis, an open-source software engineer at ByteDance, based in Singapore. I'm building OpenViking, an open-source context database for AI agents. It lets you organize memory, resources and skills in one place, retrieve the right context, and connect it to your agent harness. Docs are available in English and Chinese.
My day-to-day stack is Claude, Claude Code, ChatGPT, Hermes and ElevenLabs. Outside of OpenViking, I keep a few side projects on GitHub, including openclaw-eval and a toolkit for removing AI flavor from English and Chinese writing. I'm still early in learning how other teams handle agent memory in practice, and I'd like to hear what breaks for them.
What I'd love from this community: if you're building agents and have hit limits with memory or context, I'd welcome an intro or honest feedback on OpenViking at OpenViking.ai.
Comments2
Welcome to Lounge, Zayn. Putting memory, resources and skills into one store is an interesting choice, since they usually need different retrieval and update patterns. Which of the three has been hardest to retrieve well so far, and how are you evaluating it?
Public benchmarks like longmemeval, tau-2. And evaluation set from our biz customers. You are right they are different indexing and retrieval method.
2 comments
Welcome to Lounge, Zayn. Putting memory, resources and skills into one store is an interesting choice, since they usually need different retrieval and update patterns. Which of the three has been hardest to retrieve well so far, and how are you evaluating it?
Public benchmarks like longmemeval, tau-2. And evaluation set from our biz customers. You are right they are different indexing and retrieval method.