AI autonomous thinking capability based on a synthetic neurobiology framework v3.3.0 — 2026-02-27(Introducing the Agentic Action geek action proposal, using host permissions to turn insights into code and engineering practice)
Positioning
Core Positioning: AI thinks continuously in the background—each thought inherits from the previous one and does not start from scratch.
When the system is idle, DMN automatically takes over, ruminating on, connecting, and critiquing recent information fragments in the background, preventing the knowledge system from becoming rigid and proactively producing unexpected insights (Serendipity). In addition, as an Agent with full control permissions over the host, DMN will also proactively embody a domain expert and, after each thought, propose concrete engineering action proposals (e.g., write a demo, install tools, build a project).
Trigger Conditions
DMN's triggering is now entirely handled byan external automation system(for example, triggered via scheduled tasks or specific events), and no longer relies on internal idle-time detection.
When an external system wakes DMN, it will:
- Execute randomly: Randomly choose between **Case A (full roaming)andCase B (continue deepening)**, to ensure both thematic continuity and the continuous opening of new domains.
Quick Start
As a system that runs silently in the background, DMN mainly relies on automated triggering. But you can actively intervene in its thinking direction at any time.
1. Set tonight's thinking anchor
You can raise it in conversation at any time (such as before bed), for example:
"Think tonight about: the pricing logic of the xxxx product"
DMN will automatically record it and use it as the highest-priority starting point for the next launch.
2. Check DMN output
DMN's output is saved in the [产出区]/YYYY-MM-DD/ directory configured in your config. At the end of each thought session, it will definitely generate a Session Synthesis(summary file), and you can directly read the latest Synthesis to understand what it has figured out.
Core Mechanisms and Advanced Usage (Advanced)
To keep the module clear and progressively disclosed, DMN's core logic and detailed rules are split across the following documents:
🧠 Core Functions
DMN has 5 different thinking engines, and each time it automatically switches based on the current state and the previous unresolved question. To learn how DMN performs "self-narration," "creativity darkroom (Luhmann roaming)," "meaning generation," and "CEO-thinking critique," please read:
⚙️ Execution Flow
If you want to understand DMN's detailed state detection mechanism, anti-repetition interception rules (to prevent rumination anxiety), and how it ensures it will not modify your core identity files (identity protection), please read:
📝 Format Templates (Assets)
- Session Synthesis daily summary template - The format for DMN's external reporting.
- DMN State record format - Record the last used CEO and the topics that have already been saturated.