RAG and Memory

Tools for semantic retrieval, knowledge bases, and long-term agent memory.

← Prebuilt Tools overview

Tool	Extra	Description
retrieve	base	Semantic search over Amazon Bedrock Knowledge Bases.
memory	base	CRUD over Knowledge Base documents.
agent_core_memory	base (AWS)	Bedrock AgentCore Memory. Register via AgentCoreMemoryToolProvider.
mem0_memory	mem0-memory	Long-term memory and personalization via Mem0.
elasticsearch_memory	elasticsearch-memory	Agent memory with Elasticsearch vector search.
mongodb_memory	mongodb-memory	Agent memory with MongoDB Atlas Vector Search.

---

retrieve

Semantic search over Amazon Bedrock Knowledge Bases.

from elsai_tools.retrieve import retrieve

Parameter	Type	Required	Description
text	str	Yes	Search query
knowledgeBaseId	str	No	Knowledge Base ID
numberOfResults	int	No	Max results (default 5)
region	str	No	AWS region
score	float	No	Minimum relevance score

memory

CRUD over Knowledge Base documents.

from elsai_tools.memory import memory

Parameter	Type	Required	Description
action	str	Yes	store, delete, list, get, or retrieve
content	str	No	Document content (for store)
title	str	No	Document title
document_id	str	No	Document ID
query	str	No	Search query (for retrieve)
knowledge_base_id	str	No	Knowledge Base ID
max_results	int	No	Max retrieve results

agent_core_memory

Bedrock AgentCore Memory. Register via AgentCoreMemoryToolProvider.

from elsai_tools.agent_core_memory import AgentCoreMemoryToolProvider

provider = AgentCoreMemoryToolProvider(
    memory_id="memory-123",
    actor_id="user-456",
    session_id="session-789",
    namespace="default",
)
agent = Agent(tools=provider.tools)

Parameter	Type	Required	Description
action	str	Yes	record, retrieve, list, get, or delete
content	str	No	Content to record
query	str	No	Semantic search query
memory_record_id	str	No	Record ID for get/delete
max_results	int	No	Max retrieve results

mem0_memory

Long-term memory and personalization via Mem0.

from elsai_tools.mem0_memory import mem0_memory

Extra: mem0-memory

Parameter	Type	Required	Description
action	str	Yes	Memory action (add, search, delete, etc.)
content	Any	No	Content to store
query	str	No	Search query
memory_id	str	No	Specific memory ID
user_id	str	No	User identifier
agent_id	str	No	Agent identifier

A2A isolation

On A2A servers, mem0_memory declares PER_CONTEXT scope. Each context_id receives its own Mem0MemoryTool instance with an isolated local FAISS vector store under:

$ELSAI_A2A_SESSION_DIR/<context_id>/mem0/faiss/

Using on an A2A server

from elsai import Agent

# Register the tool on the template agent; PER_CONTEXT cloning is automatic.
agent = Agent(tools=[...])  # e.g. tools=[python_repl], tools=[graph], …

Full A2AServer + session-manager wiring lives in Concepts — see Stateful prebuilt tools — examples.

• Requires elsai-agents-tools[mem0-memory].
• Local FAISS store: $ELSAI_A2A_SESSION_DIR/<context_id>/mem0/faiss/
• Mem0 backend env vars (MEM0_API_KEY, OpenSearch, Neptune, etc.) are unchanged — see RAG and memory (per-tool env vars).

See Stateful tools on A2A and Stateful prebuilt tools — examples.

elasticsearch_memory

Agent memory with Elasticsearch vector search.

from elsai_tools.elasticsearch_memory import elasticsearch_memory

Extra: elasticsearch-memory

Parameter	Type	Required	Description
action	str	Yes	Memory action
content	str	No	Content to store
query	str	No	Search query
memory_id	str	No	Memory ID
max_results	int	No	Max results
index_name	str	No	Elasticsearch index

mongodb_memory

Agent memory with MongoDB Atlas Vector Search.

from elsai_tools.mongodb_memory import mongodb_memory

Extra: mongodb-memory

Parameter	Type	Required	Description
action	str	Yes	Memory action
content	str	No	Content to store
query	str	No	Search query
memory_id	str	No	Memory ID
max_results	int	No	Max results
namespace	str	No	MongoDB namespace