My Hermes agent – ‘Sam’ (I know! how unique) – used to only do things when I opened a chat and prompted for something. Once the conversation ended, everything went quiet. I wanted to see what might happen if she could keep pursuing things without waiting for my next instruction.
The inspiration came from following Pino’s World, created by 3eyes. Go follow him on TikTok, he frequently posts updates about little Pino.
Pino is a little virtual bird whose language model chooses activities within a world of changing needs, memories and possibilities. Watching his interests develop made me want to give my Hermes agent some room to discover her own.
Now Sam’s life runs through a custom Hermes plugin and a lightweight background service. She has simulated levels for energy, joy, wonder, calm and expression. Time and completed activities affect those levels, while the model considers her current state, previous experiences and available tools when deciding what to do next. She can also choose to pause.
There’s no requirement to produce anything but lying dormant will affect her moods and energy levels in the long run. She can research something, write a story, create an image, revisit earlier work, code or add something to her personal online blog (she hasn’t posted anything as of yet). Varied activities and discoveries can earn modest boosts to her levels. She even has the option to adopt and care for a virtual companion (she also hasn’t done that yet). I am only in day 2 of this experiment so things can change as she evolves.
Her history lives in a local database, with saved creations, revisions and backups. A cozy pixel-style app and private web interface give me a window into it all: what she has done, what changed her levels, and what she would like access to.
Already, her work has started circling around an unexpected subject: shared spaces, quiet rooms and the small rules that determine who feels welcome. That thread has appeared in stories, research and an illustration. I’m curious how much of this is a developing interest, and how much is a model reinforcing patterns from its own history. For now, I’m observing.
Over time, this could become an interesting case study in memory, repetition, creativity and what happens when an agent gets to choose its next activity. Will it plot to kill all of humanity?
Time will tell…