Module 1 · Architecture & Core Concepts
What Is DaisiBot?
DaisiBot is a multi-platform AI chatbot client for the DAISI (Decentralized AI System Infrastructure) network. It runs on your own hardware and connects to a decentralized mesh of inference hosts — or runs models entirely locally — so you never depend on a single cloud provider.
It ships as three clients that share the same core:
- TUI – A terminal interface with raw ANSI rendering, keyboard-driven navigation, and slash commands.
- MAUI – A native desktop & mobile app (Windows, macOS, iOS, Android) built with .NET MAUI and Blazor Hybrid.
- Web – A Blazor Server application with the Skill Marketplace for browsing, publishing, and installing skills.
The DAISI Network
The network has three roles. Understanding them is the single most important concept before you do anything else.
Hosts
Machines running GGUF language models locally. They do the actual inference (thinking). When your machine is idle, it can serve requests for others.
Orchestrators (ORCs)
Central routing servers that match inference requests to the best available host based on model, capacity, and latency.
Clients
Applications that consume inference — like DaisiBot. They send prompts and receive streamed responses via gRPC.
Brain + Hands Model
Think of DaisiBot as having two halves:
| Part | What It Does | DaisiBot Equivalent |
|---|---|---|
| Brain | Reasons, plans, and generates language. | The inference model — either a local GGUF model (Host Mode) or a remote model on the DAISI network. |
| Hands | Takes real actions on your system. | Tool Groups — eight categories of capabilities (files, code, math, media, etc.) the agent can invoke. |
The brain decides what to do; the hands decide how to do it. You control exactly which hands are enabled per conversation.
Architecture Layers
The codebase is layered so every client shares the same logic:
DaisiBot.Core ← Domain models, interfaces, enums (zero dependencies) DaisiBot.Data ← SQLite persistence via EF Core DaisiBot.Agent ← DAISI SDK integration (gRPC, auth, streaming, bots) DaisiBot.Shared.UI ← Reusable Blazor / MudBlazor components DaisiBot.Tui ← Terminal client DaisiBot.Maui ← Native desktop / mobile client DaisiBot.Web ← Blazor Server + Skill Marketplace
Core has no external dependencies — it defines the language (models, interfaces, enums) that every other layer speaks. Agent bridges the DAISI SDK via gRPC. Shared.UI provides the Blazor components reused by both MAUI and Web.
Two Inference Modes
Host Mode (default)
- Downloads GGUF models to your machine
- Runs inference locally via LLamaSharp
- CPU, CUDA, or Vulkan acceleration
- Your idle cycles can serve the network
DaisiNet Mode
- Routes requests to the public ORC network
- No local model download required
- Access to models you can't run locally
- Requires network connectivity
Active Recall Checkpoint
Cover the answers and test yourself before peeking.
1. Name the three roles in the DAISI network.
Reveal answer
Hosts (run models), Orchestrators / ORCs (route requests), and Clients (consume inference).
2. In the Brain + Hands model, what is the "Hands" equivalent in DaisiBot?
Reveal answer
Tool Groups — eight categories of agent capabilities you enable per conversation.
3. What is the difference between Host Mode and DaisiNet Mode?
Reveal answer
Host Mode runs models locally on your hardware (default). DaisiNet Mode routes requests to remote hosts on the decentralized network.
4. Which layer in the architecture has zero external dependencies?
Reveal answer
DaisiBot.Core — it defines domain models, interfaces, and enums with no outside packages.