Session is Agent: Monolithic Architecture Philosophy
UNIChat adopts a unique "monolithic" design philosophy: Session is Agent.
Traditional Model vs UNIChat Model
| Feature | Traditional Chat Tools | UNIChat |
|---|---|---|
| Configuration Scope | Global or Character Library | Deeply bound within the current session |
| Knowledge Base | Global association | Session-private association |
| Flexibility | Modifying a role affects all related chats | Each conversation is an independently evolving entity |
| Complexity | High (requires jumping between multiple interfaces) | Low (One-stop configuration) |
Why Design It This Way?
1. Reduce Context Pollution
If you set up a "Coding Assistant" role globally, but you want him to switch to the Python 3.12 standard library in a specific conversation, traditional tools might require you to modify global settings. In UNIChat, you only need to fine-tune it in the current session's settings, without interfering with other conversations.
2. State Integrity
An Agent should encompass its:
- Persona: Who is it?
- Knowledge: What does it know?
- Parameters: How is its reply style (Temperature, Top-P, etc.)?
Locking all these attributes within the session means that when you export or share this session, you are sharing a complete, reproducible state of an intelligent entity.
Operation Guide
By clicking [⚙️ Agent Settings] at the top of the session, you can adjust at any time:
- ⚙️ Configuring Models and Parameters: This is the Agent's "brain". Choose a suitable model for it (like DeepSeek or GPT-4) and configure the basic parameter details.
- 🗣️ Writing System Prompts: This is the Agent's "personality". Use prompts to define who it is, what it is responsible for, and how it communicates with you.
- 👤 Binding User Identity: This is how the Agent recognizes you. You can assign a specific persona to it or have it use a special "nickname" for you.