RAG and memory

Optional extras:

pip install --extra-index-url https://elsai-agents.elsai.ai/root/ "elsai-agents-tools[mem0-memory]==0.3.0"
pip install --extra-index-url https://elsai-agents.elsai.ai/root/ "elsai-agents-tools[elasticsearch-memory]==0.3.0"
pip install --extra-index-url https://elsai-agents.elsai.ai/root/ "elsai-agents-tools[mongodb-memory]==0.3.0"

Tools

retrieve

Query Amazon Bedrock Knowledge Bases with semantic search and return ranked passages for agent context

Environment variables

Variable	Required	Default	Description
KNOWLEDGE_BASE_ID	yes	—	Default Bedrock Knowledge Base ID for retrieve
AWS_REGION	yes	—	AWS region for Bedrock / AgentCore (needed when not already configured in your AWS environment)
MIN_SCORE	optional	0.4	Lowest relevance score to keep from retrieve
RETRIEVE_ENABLE_METADATA_DEFAULT	optional	false	Whether retrieve results include metadata by default

memory

Full CRUD over Knowledge Base documents — store, list, get, delete, and retrieve entries the agent can reference later

Environment variables

Variable	Required	Default	Description
ELSAI_KNOWLEDGE_BASE_ID	yes	—	Default Knowledge Base ID for memory (separate from KNOWLEDGE_BASE_ID)
AWS_REGION	yes	—	AWS region for Bedrock / AgentCore (needed when not already configured in your AWS environment)
MEMORY_DEFAULT_MIN_SCORE	optional	0.4	Lowest relevance score for memory retrieve
MEMORY_DEFAULT_MAX_RESULTS	optional	50	Max results for memory retrieve

Also respects BYPASS_TOOL_CONSENT (see Tool consent).

agent_core_memory

Persistent short- and long-term memory through Bedrock AgentCore — record facts, retrieve by query, and manage session history

Extra: base (AWS)

Environment variables

Variable	Required	Default	Description
AWS_REGION	yes	—	AWS region for Bedrock / AgentCore (needed when not already configured in your AWS environment)

mem0_memory

Long-term agent memory and user personalization backed by Mem0, with vector search over past interactions

Extra: mem0-memory

Environment variables

Variable	Required	Default	Description
MEM0_API_KEY	yes	—	Mem0 cloud API key (uses the cloud backend when set)
OPENSEARCH_HOST	yes	—	OpenSearch host for Mem0 (do not combine with Neptune Analytics graph ID)
NEPTUNE_ANALYTICS_GRAPH_IDENTIFIER	yes	—	Neptune Analytics graph for Mem0 vector storage
NEPTUNE_DATABASE_ENDPOINT	yes	—	Neptune DB endpoint for Mem0
AWS_REGION	yes	—	AWS region for Bedrock / AgentCore (needed when not already configured in your AWS environment)
MEM0_EMBEDDER_PROVIDER	optional	aws_bedrock	Embedder provider for Mem0
MEM0_EMBEDDER_MODEL	optional	amazon.titan-embed-text-v2:0	Embedding model for Mem0
MEM0_LLM_PROVIDER	optional	aws_bedrock	LLM provider for Mem0
MEM0_LLM_MODEL	optional	Claude Haiku default	LLM model for Mem0
MEM0_LLM_TEMPERATURE	optional	0.1	LLM temperature for Mem0
MEM0_LLM_MAX_TOKENS	optional	2000	LLM max tokens for Mem0
OPENSEARCH_COLLECTION	optional	mem0	OpenSearch collection name
NEPTUNE_ANALYTICS_VECTOR_COLLECTION	optional	mem0	Neptune Analytics vector collection name

Also respects BYPASS_TOOL_CONSENT (see Tool consent).

elasticsearch_memory

Store and recall agent memory in Elasticsearch with vector similarity search for fast semantic lookup

Extra: elasticsearch-memory

Environment variables

Variable	Required	Default	Description
ELASTICSEARCH_URL / ELASTICSEARCH_CLOUD_ID	yes	—	Elasticsearch connection URL or cloud ID
ELASTICSEARCH_API_KEY	yes	—	Elasticsearch API key (when your cluster requires it)
ELASTICSEARCH_INDEX_NAME	yes	—	Elasticsearch index for agent memory

mongodb_memory

Store and retrieve agent memory using MongoDB Atlas Vector Search for scalable, document-oriented recall

Extra: mongodb-memory

Environment variables

Variable	Required	Default	Description
MONGODB_ATLAS_CLUSTER_URI	yes	—	MongoDB Atlas connection URI
MONGODB_DATABASE_NAME	yes	—	MongoDB database name
MONGODB_COLLECTION_NAME	yes	—	MongoDB collection name

Examples

retrieve

Requires OPENAI_API_KEY, AWS credentials, and KNOWLEDGE_BASE_ID.

pip install --extra-index-url https://elsai-agents.elsai.ai/root/ elsai-agents==0.3.5
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-model/ "elsai-model[openai]==2.1.1"
pip install --extra-index-url https://elsai-agents.elsai.ai/root/ elsai-agents-tools==0.3.0

import os

from elsai import Agent
from elsai.agent import AgentConfig
from elsai_model.openai import OpenAIModel
from elsai_tools.retrieve import retrieve

agent = Agent(
    model=OpenAIModel(
        model_id="gpt-4o-mini",
        client_args={"api_key": os.environ["OPENAI_API_KEY"]},
    ),
    tools=[retrieve],
    system_prompt=(
        "Use retrieve to search the knowledge base before answering. "
        "After using tools, answer the user clearly in plain text."
    ),
    config=AgentConfig(name="rag_assistant"),
)

result = agent("What does our knowledge base say about refunds?")
print(result)

mem0_memory

Requires OPENAI_API_KEY, MEM0_API_KEY, and the Mem0 extra. For demos set BYPASS_TOOL_CONSENT=true.

pip install --extra-index-url https://elsai-agents.elsai.ai/root/ elsai-agents==0.3.5
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-model/ "elsai-model[openai]==2.1.1"
pip install --extra-index-url https://elsai-agents.elsai.ai/root/ "elsai-agents-tools[mem0-memory]==0.3.0"

import os

from elsai import Agent
from elsai.agent import AgentConfig
from elsai_model.openai import OpenAIModel
from elsai_tools.mem0_memory import mem0_memory

agent = Agent(
    model=OpenAIModel(
        model_id="gpt-4o-mini",
        client_args={"api_key": os.environ["OPENAI_API_KEY"]},
    ),
    tools=[mem0_memory],
    system_prompt=(
        "Use mem0_memory to store and recall user preferences. "
        "After using tools, answer the user clearly in plain text."
    ),
    config=AgentConfig(name="memory_assistant"),
)

result = agent("Remember that my favourite colour is blue, then confirm what you stored.")
print(result)

API reference

See RAG and memory API for parameters and import paths.

See also

• Prebuilt Tools overview