Overview of AutoGRAMS concepts
-
AutoGRAMS basics -- gives an introduction to the main features and necessary concepts needed to understand and use AutoGRAMS. It includes:
- Initializing the autogram -- shows how to load or define an autogram in python
- Getting replies -- shows how to get a reply from an initialized autogram chatbot
- Nodes -- all autograms are composed of a set of nodes which the autogram's designer must specify.
- Transitions -- Each node applies a transition to select another node after executing. Several fields define how the node selects which node will come next
- Instructions -- Each node has an
instruction
field that in combination with theaction
field define what happens when the node executes. Depending on the action, the instruction may for instance be interpreted as a language model prompt or Python code to execute. - Internal functions and scopes -- describes how to execute AutoGRAMS graphs as callable modules
- Calling AutoGRAMS functions from python -- describes how AutoGRAMS graphs designed as callable modules can be called directly from Python.
- Python functions -- describes how Python code can be embedded within an AutoGRAMS node
- Variables -- describes how nodes can be used to set variables in memory, and how to reference these variables to use them later.
- Memory Object -- introduces how memory is managed in AutoGRAMS
-
Nodes and actions --
action
is a node field that define how the node's instruction will be interpreted, and determines what type of node it is. This page describes all the actions available in AutoGRAMS. -
Transitions -- Detailed guide transitions in AutoGRAMS that describes them in more depth and gives more examples
-
Interjection Nodes -- Guide to a special type of node called an "interjection node "that allows transitions from any other chat node.
-
Autogram Config -- describes Autogram config, a python class used to define general agent settings such as what language models to use or what prompt template to use.
-
Prompt formation -- describes how prompts are formed in AutoGRAMS, using a combination of the language model turn history and the node specific instructions
-
Models and generation - describes how models are used to generate text for replies and classify multiple choice questions for transitions
-
AutoGRAMS Compiled from Python -- describes how to code autograms using python like syntax. This method of implementation is convenient for integrating Python code more deeply with AutoGRAMS nodes.
-
Node Subtypes -- A detailed overview of the class structure and behavior of node subtypes, which are determined by a node's action when it is defined.
-
Understanding the interpreter -- Explains how autograms are executed under the hood
-
Autogram Compiler -- Explains how python-like code in AutoGRAMS compiled from Python is converted to an set of AutoGRAMS nodes using the Python abstract syntax tree
-
Glossary of node fields -- Describes all the different fields that can be set for AutoGRAMS nodes