Experimental Modules#

This module contains experimental modules and reproductions of existing work using LangChain primitives.

Autonomous Agents#

Here, we document the BabyAGI and AutoGPT classes from the langchain.experimental module.

class langchain.experimental.BabyAGI(*, memory: Optional[langchain.schema.BaseMemory] = None, callbacks: Optional[Union[List[langchain.callbacks.base.BaseCallbackHandler], langchain.callbacks.base.BaseCallbackManager]] = None, callback_manager: Optional[langchain.callbacks.base.BaseCallbackManager] = None, verbose: bool = None, task_list: collections.deque = None, task_creation_chain: langchain.chains.base.Chain, task_prioritization_chain: langchain.chains.base.Chain, execution_chain: langchain.chains.base.Chain, task_id_counter: int = 1, vectorstore: langchain.vectorstores.base.VectorStore, max_iterations: Optional[int] = None)[source]#

Controller model for the BabyAGI agent.

model Config[source]#

Configuration for this pydantic object.

arbitrary_types_allowed = True#

execute_task(objective: str, task: str, k: int = 5) → str[source]#: Execute a task.

classmethod from_llm(llm: langchain.base_language.BaseLanguageModel, vectorstore: langchain.vectorstores.base.VectorStore, verbose: bool = False, task_execution_chain: Optional[langchain.chains.base.Chain] = None, **kwargs: Dict[str, Any]) → langchain.experimental.autonomous_agents.baby_agi.baby_agi.BabyAGI[source]#: Initialize the BabyAGI Controller.

get_next_task(result: str, task_description: str, objective: str) → List[Dict][source]#: Get the next task.

property input_keys: List[str]#: Input keys this chain expects.

property output_keys: List[str]#: Output keys this chain expects.

prioritize_tasks(this_task_id: int, objective: str) → List[Dict][source]#: Prioritize tasks.

class langchain.experimental.AutoGPT(ai_name: str, memory: langchain.vectorstores.base.VectorStoreRetriever, chain: langchain.chains.llm.LLMChain, output_parser: langchain.experimental.autonomous_agents.autogpt.output_parser.BaseAutoGPTOutputParser, tools: List[langchain.tools.base.BaseTool], feedback_tool: Optional[langchain.tools.human.tool.HumanInputRun] = None)[source]#: Agent class for interacting with Auto-GPT.

Generative Agents#

Here, we document the GenerativeAgent and GenerativeAgentMemory classes from the langchain.experimental module.

class langchain.experimental.GenerativeAgent(*, name: str, age: Optional[int] = None, traits: str = 'N/A', status: str, memory: langchain.experimental.generative_agents.memory.GenerativeAgentMemory, llm: langchain.base_language.BaseLanguageModel, verbose: bool = False, summary: str = '', summary_refresh_seconds: int = 3600, last_refreshed: datetime.datetime = None, daily_summaries: List[str] = None)[source]#

A character with memory and innate characteristics.

model Config[source]#

Configuration for this pydantic object.

arbitrary_types_allowed = True#

field age: Optional[int] = None#: The optional age of the character.

field daily_summaries: List[str] [Optional]#: Summary of the events in the plan that the agent took.

generate_dialogue_response(observation: str, now: Optional[datetime.datetime] = None) → Tuple[bool, str][source]#: React to a given observation.

generate_reaction(observation: str, now: Optional[datetime.datetime] = None) → Tuple[bool, str][source]#: React to a given observation.

get_full_header(force_refresh: bool = False, now: Optional[datetime.datetime] = None) → str[source]#: Return a full header of the agent’s status, summary, and current time.

get_summary(force_refresh: bool = False, now: Optional[datetime.datetime] = None) → str[source]#: Return a descriptive summary of the agent.

field last_refreshed: datetime.datetime [Optional]#: The last time the character’s summary was regenerated.

field llm: langchain.base_language.BaseLanguageModel [Required]#: The underlying language model.

field memory: langchain.experimental.generative_agents.memory.GenerativeAgentMemory [Required]#: The memory object that combines relevance, recency, and ‘importance’.

field name: str [Required]#: The character’s name.

field status: str [Required]#: The traits of the character you wish not to change.

summarize_related_memories(observation: str) → str[source]#: Summarize memories that are most relevant to an observation.

field summary: str = ''#: Stateful self-summary generated via reflection on the character’s memory.

field summary_refresh_seconds: int = 3600#: How frequently to re-generate the summary.

field traits: str = 'N/A'#: Permanent traits to ascribe to the character.

class langchain.experimental.GenerativeAgentMemory(*, llm: langchain.base_language.BaseLanguageModel, memory_retriever: langchain.retrievers.time_weighted_retriever.TimeWeightedVectorStoreRetriever, verbose: bool = False, reflection_threshold: Optional[float] = None, current_plan: List[str] = [], importance_weight: float = 0.15, aggregate_importance: float = 0.0, max_tokens_limit: int = 1200, queries_key: str = 'queries', most_recent_memories_token_key: str = 'recent_memories_token', add_memory_key: str = 'add_memory', relevant_memories_key: str = 'relevant_memories', relevant_memories_simple_key: str = 'relevant_memories_simple', most_recent_memories_key: str = 'most_recent_memories', now_key: str = 'now', reflecting: bool = False)[source]#

add_memories(memory_content: str, now: Optional[datetime.datetime] = None) → List[str][source]#: Add an observations or memories to the agent’s memory.

add_memory(memory_content: str, now: Optional[datetime.datetime] = None) → List[str][source]#: Add an observation or memory to the agent’s memory.

field aggregate_importance: float = 0.0#

Track the sum of the ‘importance’ of recent memories.

Triggers reflection when it reaches reflection_threshold.

clear() → None[source]#: Clear memory contents.

field current_plan: List[str] = []#: The current plan of the agent.

fetch_memories(observation: str, now: Optional[datetime.datetime] = None) → List[langchain.schema.Document][source]#: Fetch related memories.

field importance_weight: float = 0.15#: How much weight to assign the memory importance.

field llm: langchain.base_language.BaseLanguageModel [Required]#: The core language model.

load_memory_variables(inputs: Dict[str, Any]) → Dict[str, str][source]#: Return key-value pairs given the text input to the chain.

field memory_retriever: langchain.retrievers.time_weighted_retriever.TimeWeightedVectorStoreRetriever [Required]#: The retriever to fetch related memories.

property memory_variables: List[str]#: Input keys this memory class will load dynamically.

pause_to_reflect(now: Optional[datetime.datetime] = None) → List[str][source]#: Reflect on recent observations and generate ‘insights’.

field reflection_threshold: Optional[float] = None#: When aggregate_importance exceeds reflection_threshold, stop to reflect.

save_context(inputs: Dict[str, Any], outputs: Dict[str, Any]) → None[source]#: Save the context of this model run to memory.

Experimental Modules

Contents

Experimental Modules#

Autonomous Agents#

Generative Agents#