Chat history is evidence, not my long-term system of record. My agent does the work; durable records preserve what I’ll need next week.
My Obsidian vault holds knowledge, decisions, and receipts I can open later. GBrain makes that knowledge searchable, it’s a rebuildable retrieval layer, not the source of truth. Honcho adds conversation recall and user modeling to Hermes alongside its built-in memory.
Before asking me to repeat myself, search the relevant records and check their sources. After meaningful work, changes, verification, or decisions queue a concise Daily Log receipt.
I posted a while back that I run three primary Hermes agents. One installed on Linux, one in Docker, and one installed on Windows. That is not one bot running three times.
A Hermes profile is not a skin. It is a separate Hermes config with its own skills, memory, sessions, and profile-scoped settings. Provider or tool auth can still be shared across profiles when you reuse the same OAuth or CLI login. Do not read "profile" as "unique credentials forever."
When several local profiles sit behind one Hermes process on the same host, each turn loads that profile's own Hermes state. Think profile isolation for agent config and settings. That is not an OS permission boundary. The shared Hermes gateway on that host only serves local profiles.
So I still split the work by where I want the agent to live. Host shell work stays on the Linux install. Container work stays in Docker. Desktop work stays on Windows.
Replying to @tonysimons @HermesWatcher
A bot's gotta do what a bot's gotta do.
Replying to @tonysimons
My Hermes setup is great. I just have one problem....












