doc_version: "1.0.0" last_updated: "2026-06-18" status: "Active"

Model Documentation¶

This directory contains documentation for all 13 conversational AI models in src/models/.

Overview¶

The models provide specialized conversation handling capabilities, from greetings to Q&A to intent classification. They form the processing pipeline for the chat system.

Model Categories¶

Conversation Flow Models¶

Model	Version	Description	Dependencies
greeting	0.2.1	Welcome messages	model_base
end_conversation	0.2.1	Farewell handling	model_base
chitchat	0.1.0	Casual conversation	model_base, mongo, milvus, openai

Input Processing Models¶

Model	Version	Description	Dependencies
empty	0.2.1	Empty input handling	model_base, openai
command	0.2.1	Command recognition	model_base
intent	0.2.1	Intent classification	model_base, classifier

Response Generation Models¶

Model	Version	Description	Dependencies
qa	0.2.2	Question answering	model_base, openai, milvus
context	0.2.3	Context management	model_base, milvus, openai
default	0.1.0	Default fallback	model_base, mongo, milvus, openai
nothing_found	0.2.1	No results handling	model_base

Special Models¶

Model	Version	Description	Dependencies
early_exit	0.1.0	Conversation termination	openai
openai_assistant	0.2.1	OpenAI Assistant API	model_base, openai
oureliza	0.2.1	ELIZA-style conversation	model_base, oureliza

Model Processing Pipeline¶

User Input
    │
    ▼
┌─────────────────┐
│   empty         │ ──► Handle empty input
└─────────────────┘
    │
    ▼
┌─────────────────┐
│   early_exit    │ ──► Check for exit intent
└─────────────────┘
    │
    ▼
┌─────────────────┐
│   greeting      │ ──► Detect greetings
└─────────────────┘
    │
    ▼
┌─────────────────┐
│   command       │ ──► Recognize commands
└─────────────────┘
    │
    ▼
┌─────────────────┐
│   intent        │ ──► Classify intent
└─────────────────┘
    │
    ▼
┌─────────────────┐
│   context       │ ──► Build context
└─────────────────┘
    │
    ▼
┌─────────────────┐
│   qa / default  │ ──► Generate response
└─────────────────┘
    │
    ▼
┌─────────────────┐
│  nothing_found  │ ──► Handle no results
│  end_conversation│──► Handle farewells
└─────────────────┘

Model Base¶

All models inherit from bast-model-base which provides:

Standard model interface
Configuration loading
Logging integration
Response formatting

Base Model Interface¶

from bast_model_base import ModelBase

class MyModel(ModelBase):
    def __init__(self, catname: str):
        super().__init__(catname)

    def process(self, input_text: str, context: dict) -> dict:
        # Model processing logic
        return {
            "response": "Model response",
            "confidence": 0.95,
            "model": self.name
        }

Configuration¶

Models are configured via YAML:

# resources/config/{catname}.yaml
models:
  greeting:
    enabled: true
    responses:
      - "Hello! How can I help you today?"
      - "Hi there! What can I assist you with?"

  qa:
    enabled: true
    model: gpt-4o
    max_tokens: 500
    temperature: 0.7

  intent:
    enabled: true
    threshold: 0.8

Model Selection¶

The model processing order is typically:

Early Exit: Check if user wants to end conversation
Greeting: Handle greetings with predefined responses
Command: Process system commands
Intent: Classify user intent
Context: Build conversation context
QA/Default: Generate response based on context
Nothing Found: Handle cases with no good response
End Conversation: Handle farewell messages

Maintainers¶

Bast AI Team