At first I had a functional agent running on Hermes and local models (DeepSeek v4 flash) via vllm and a dual DGX Spark cluster. My wife and kids wanted to use it so I build Cinnamon to have personal and shared contexts (and separate facts and memory depending on the context).
I think it would be great as open source so families can manage data privacy because everyone is building one of these now. The page looked really similar to another one I saw recently: https://heydana.ai/
Your implementation of family vs private space is nice, but I think now that there really isn't a moat except maybe privacy/control.
It's not the control around the memories, but that without this being open source/self operable, I can't tell what it's doing with the requests that generated those memories.
Logging and telemetry, for example: could it leak info? What are the controls around access to the info? These things are questions I think about if I'm going to put my kids on a system like this. Put yourself on the other side and why you would use your own agent with your family, but not Dana (example): because you want full control over the data.
Thank you for sharing the link on Dana, I haven't heard of it before. There's no technological moat anymore, only marketing. It's still fun to compete in this space because of how easy it is to quickly iterate.
At first I had a functional agent running on Hermes and local models (DeepSeek v4 flash) via vllm and a dual DGX Spark cluster. My wife and kids wanted to use it so I build Cinnamon to have personal and shared contexts (and separate facts and memory depending on the context).
I think it would be great as open source so families can manage data privacy because everyone is building one of these now. The page looked really similar to another one I saw recently: https://heydana.ai/
Your implementation of family vs private space is nice, but I think now that there really isn't a moat except maybe privacy/control.
Privacy & control are huge. I did two things:
1. made facts and memories easily accessible (and you can view, correct, or delete what Cinnamon "knows" about you in a given context)
2. observable execution flow with multiple guardrails (and one subagent focused entirely on data safety)
Happy to show screenshots if anyone's interested.
It's not the control around the memories, but that without this being open source/self operable, I can't tell what it's doing with the requests that generated those memories.
Logging and telemetry, for example: could it leak info? What are the controls around access to the info? These things are questions I think about if I'm going to put my kids on a system like this. Put yourself on the other side and why you would use your own agent with your family, but not Dana (example): because you want full control over the data.
This is how I am doing it currently:
https://imgur.com/a/KYTAJGU
Would love to hear your recommendations.
[dead]
Thank you for sharing the link on Dana, I haven't heard of it before. There's no technological moat anymore, only marketing. It's still fun to compete in this space because of how easy it is to quickly iterate.
[flagged]