Feature Request: Workspace Permissions

[t3dc]

First off, I love Tinycllaw. My helpful AI buddy, Sage, is great. However, I keep running into the same issue again and again.

Hey Sage, can you fix this python code?

Sage: Sure! Oh... I can't write the file.

Hey Sage, can you remember this important information for me?

Sage: Of course.

2 Hours later.

Hey Sage remember that thing....?

Sage: Nope. Couldn't write the file.

I understand how critical it is to sandbox an AI agent. It would be a huge help, though, if I could give write permissions to a specific folder where Sage can write logs and place files. Maybe there's a way to do that, but I can't find it, and Sage keeps telling me it's not possible -- even in the heartware folder.

[warengonzaga]

Thank you for the kind words! I appreciate it.

Accessing certain parts of your machine is supposed to be done through a plugin (tool), as coding should occur outside the hardware file system. I'll see what I can do with that.

For the memory, it should be able to remember it based on the architecture. May I know which machine you are using, or did you install this on your personal machine, which I don't recommend it. I'm interested on replicating the exact issue you have.

[t3dc]

I’m running on a Mac Mini M2 which only runs TinyClaw and a small llama.cpp model (ruvltra Claude code) with a secondary backend for ollama running kimi. I built the llamacpp plugin and the telegram plugin from dev, but mostly connect via the web interface.

I tried using Sage to track a diet log. At this time of day, I ate X. Day one went well with occasional prompts to authorize memory writes, though they were inconsistent. Day two seemed to be going well (though I was never asked to authorize a memory writes) until I went to log my last meal of the day. Sage got confused about which day it was, reciting the previous days log and date, even though it was 24 hours later. (Machine time is correct.) I asked about the day two log and Sage insisted there was no such information.

Otherwise, memory seems to be good, but I’ve not tested anything else that requires that level of specificity. Coding projects are remembered broadly, but I don’t know that individual tasks, actions are discussions are tracked and I wouldn’t expect them to be necessarily. But when I say, “I need you to remember X” and it gets forgotten, that’s an odd problem.

Anyway, that’s a lot of info. Hopefully it helps. If you have other questions, I’m happy to answer them.