Serialization

Adapted from the Griptape AI Framework documentation.

Overview

Many components in Gen AI Builder can be serialized and deserialized using the to_dict and from_dict methods. There are also to_json and from_json as a convenience.

Here is how we can serialize an Agent and then deserialize it back:

Code
Logs

from griptape.structures import Agent

agent = Agent()
agent.run("My name is Collin")

agent_dict = agent.to_dict()

new_agent = Agent.from_dict(agent_dict)

new_agent.run("What's my name?")

[03/12/25 20:36:02] INFO     PromptTask d9e3f73477ef4a0aa71a670e543ee73c        
                             Input: My name is Collin                           
                    INFO     PromptTask d9e3f73477ef4a0aa71a670e543ee73c        
                             Output: Hello, Collin! How can I assist you today? 
[03/12/25 20:36:03] INFO     PromptTask d9e3f73477ef4a0aa71a670e543ee73c        
                             Input: What's my name?                             
[03/12/25 20:36:05] INFO     PromptTask d9e3f73477ef4a0aa71a670e543ee73c        
                             Output: Your name is Collin. How can I help you    
                             further?

Serialization Overrides

All classes that implement the SerializableMixin can be serialized using the above methods. However, only fields marked with metadata={"serializable": True} will be included in the serialization process. If you need to add or remove fields in the serialization process, you can pass serialization_overrides to any of the serialization methods.

Code
Logs

from rich.pretty import pprint

from griptape.structures import Agent

agent = Agent()
agent.run("My name is Collin")

agent_dict = agent.to_dict(serializable_overrides={"max_meta_memory_entries": True, "id": False})
# `max_meta_memory_entries` will be included, `id` will not
pprint(agent_dict)

[03/12/25 20:35:19] INFO     PromptTask 4cb3fb0cb25b4fb5b9c1ec9fcc17789c        
                             Input: My name is Collin                           
[03/12/25 20:35:20] INFO     PromptTask 4cb3fb0cb25b4fb5b9c1ec9fcc17789c        
                             Output: Hello, Collin! How can I assist you today? 
{
│   'type': 'Agent',
│   'rulesets': [],
│   'rules': [],
│   'conversation_memory': {
│   │   'type': 'ConversationMemory',
│   │   'runs': [
│   │   │   {
│   │   │   │   'type': 'Run',
│   │   │   │   'id': '4ea9601351f542b6a403c82ffcf77656',
│   │   │   │   'meta': None,
│   │   │   │   'input': {
│   │   │   │   │   'type': 'TextArtifact',
│   │   │   │   │   'id': '80d96cb6f76e4d89925182d8b4e3acc7',
│   │   │   │   │   'reference': None,
│   │   │   │   │   'meta': {},
│   │   │   │   │   'name': '80d96cb6f76e4d89925182d8b4e3acc7',
│   │   │   │   │   'value': 'My name is Collin'
│   │   │   │   },
│   │   │   │   'output': {
│   │   │   │   │   'type': 'TextArtifact',
│   │   │   │   │   'id': '501e6f2dfa3d496685ce9249bdb8121c',
│   │   │   │   │   'reference': None,
│   │   │   │   │   'meta': {'is_react_prompt': False},
│   │   │   │   │   'name': '501e6f2dfa3d496685ce9249bdb8121c',
│   │   │   │   │   'value': 'Hello, Collin! How can I assist you today?'
│   │   │   │   }
│   │   │   }
│   │   ],
│   │   'meta': {},
│   │   'max_runs': None
│   },
│   'conversation_memory_strategy': 'per_structure',
│   'max_meta_memory_entries': 20,
│   'tasks': [
│   │   {
│   │   │   'type': 'PromptTask',
│   │   │   'rulesets': [],
│   │   │   'rules': [],
│   │   │   'id': '4cb3fb0cb25b4fb5b9c1ec9fcc17789c',
│   │   │   'state': 'State.FINISHED',
│   │   │   'parent_ids': [],
│   │   │   'child_ids': [],
│   │   │   'max_meta_memory_entries': 20,
│   │   │   'context': {},
│   │   │   'prompt_driver': {
│   │   │   │   'type': 'OpenAiChatPromptDriver',
│   │   │   │   'temperature': 0.1,
│   │   │   │   'max_tokens': None,
│   │   │   │   'stream': False,
│   │   │   │   'extra_params': {},
│   │   │   │   'base_url': None,
│   │   │   │   'organization': None,
│   │   │   │   'model': 'gpt-4o',
│   │   │   │   'user': '',
│   │   │   │   'response_format': None,
│   │   │   │   'seed': None,
│   │   │   │   'reasoning_effort': 'medium',
│   │   │   │   'use_native_tools': True,
│   │   │   │   'structured_output_strategy': 'native',
│   │   │   │   'parallel_tool_calls': True,
│   │   │   │   'modalities': [],
│   │   │   │   'audio': {'voice': 'alloy', 'format': 'pcm16'}
│   │   │   },
│   │   │   'tools': [],
│   │   │   'max_subtasks': 20
│   │   }
│   ]
}

Types Overrides

Due to some unfortunate internals of the Gen AI Builder's serialization process, you may occasionally run into a NameError when serializing. It will look something like this:

NameError: name 'BaseWebSearchDriver' is not defined

This is something we're actively working on fixing, but in the meantime, you can use the types_overrides parameter to pass in a dictionary of types that need to be overridden during serialization.

Here is an example of how you can use types_overrides:

Code
Logs

from duckduckgo_search import DDGS
from rich.pretty import pprint

from griptape.drivers.web_search import BaseWebSearchDriver
from griptape.drivers.web_search.duck_duck_go import DuckDuckGoWebSearchDriver
from griptape.tools import WebSearchTool

web_driver = DuckDuckGoWebSearchDriver()
web_tool = WebSearchTool(web_search_driver=web_driver)

web_tool_dict = web_tool.to_dict(
    serializable_overrides={"web_search_driver": True},
    types_overrides={"BaseWebSearchDriver": BaseWebSearchDriver, "DDGS": DDGS},
)

pprint(web_tool_dict)

{
│   'type': 'WebSearchTool',
│   'name': 'WebSearchTool',
│   'input_memory': None,
│   'output_memory': None,
│   'install_dependencies_on_init': True,
│   'dependencies_install_directory': None,
│   'verbose': False,
│   'off_prompt': False,
│   'web_search_driver': {'type': 'DuckDuckGoWebSearchDriver'}
}

← Prev

Tokenizers

↑ Up

Events

SDK reference

Could this page be better? Report a problem or suggest an addition!