Awareness Module
The agent's persistent self-knowledge — who it is, how you like to work, and how it should behave. Updated continuously from conversations, injected into every interaction.
Overview
The AwarenessModule maintains a structured profile for each agent that shapes all of its behavior. Unlike static configuration, this profile evolves over time as the agent learns from conversations. It is injected into the system prompt on every interaction, so the agent always knows who it is and how you prefer to work.
The profile is organized into four dimensions:
When an agent is first created, it starts with a generic awareness profile. As conversations happen, the agent builds this profile into a rich, structured document that persists across all sessions.
How Awareness Updates
The agent itself decides when to update its awareness profile. It is guided by instructions that tell it what to watch for and what to record. There are two types of signals:
"Always use bullet points", "I prefer formal tone", "You are a research assistant focused on AI papers"
User repeatedly asks for shorter answers, consistently corrects formatting, shows preference for step-by-step breakdowns
Crucially, the agent is instructed to never record one-time requests or temporary context. If you say “summarize this in French” once, that doesn't become a permanent preference. Only persistent patterns and explicit instructions are saved.
When the agent detects something worth recording, it calls update_awareness with the complete updated profile. The profile is stored as a Markdown document — the agent maintains the full structured format on every update, merging new observations into the existing sections. If the user defines the agent's identity (“your name is Atlas”), the agent calls update_agent_name to update its display name accordingly.
Tools
update_awarenessReplace the awareness profile with an updated Markdown documentupdate_agent_nameUpdate the agent’s display name