Redis Agent Memory Python SDK quickstart

Explore session memory, automatic extraction, summarization, custom memory types, and sensitive-data exclusions with the Redis Agent Memory Python SDK.

Use this quickstart to follow a travel planning conversation through Redis Agent Memory. You will retrieve the conversation from session memory, recall information extracted in the background, inspect an automatically generated session summary, extract structured travel information, and guide extraction away from sensitive data.

Before you begin

To complete this quickstart, you need:

• A Redis Cloud account that can create Redis Agent Memory services.
• An eligible Redis Cloud database, or permission to create one.

The database must be active, run a supported Redis version, and use a Pro or Essentials plan. It must also have a public endpoint, Redis Search, and the default user enabled.

Redis Agent Memory doesn't support Flex, Active-Active, or AWS PrivateLink databases during public preview.

For the complete list, see prerequisites and limitations.

You also need Python 3.10 or later.

Create a Redis Agent Memory service

1. Sign in to the Redis Cloud console.

2. Select Agent Memory from the navigation menu.

3. If Redis Cloud displays the public preview terms, review and accept them.

4. Select Create custom service.

5. Enter a service name, select an eligible database, and select its default user.

6. Under Memory configuration, enter these values:

   Setting	Value	What it controls
   Short-term TTL	1 day	How long session memory is retained.
   Long-term TTL	365 days	How long long-term memories are retained.
   Extraction cadence	1 minute	How often session events are processed for extraction. One minute is for this quickstart; use a longer production interval unless you need rapid extraction.
   Automatic summarization	Enabled	Whether older session events are condensed into a summary.
   Summarize after (messages)	6	The event count that triggers summarization. Six is for this quickstart; use a higher production threshold.
   Keep most recent (messages)	2	How many recent events remain in full. Two is for this quickstart; retain more in production when recent turns are needed.

7. Under Memory types & extraction, select Add type and configure this custom memory type:

   Setting	Value	What it controls
   Name	trip_preference	The identifier stored in memoryType and used in search filters.
   Description	Structured requirements for a planned trip	The purpose of the custom memory type.
   Extraction prompt	Extract trip requirements only when the user states a destination or travel plan. Preserve explicit dietary requirements and food preferences.	When to create the memory and which information to capture.
   Enabled	Enabled	Whether new memories of this type are extracted.

8. Add these custom fields:

   Field	Type	Description
   destinations	list[str]	Cities or countries the user plans to visit.
   travel_period	str	When the user plans to travel.
   dietary_requirements	list[str]	Dietary requirements that affect recommendations.
   food_preferences	list[str]	Cuisines, flavors, or dining preferences stated by the user.

9. Under Sensitive-data exclusions, enable Semantic exclusions and enter this exclusion prompt:

   Do not keep passwords, access tokens, recovery codes, payment card information, or booking confirmation codes in long-term memory.
   

10. Select Create.

11. Copy the Redis Agent Memory API key and store it securely.

These settings keep the background stages short enough to observe during the quickstart. For screenshots and configuration details, see create a Redis Agent Memory service.

Save the connection values

1. Open the Redis Agent Memory service in the Redis Cloud console.

2. On the Configuration tab, copy the Endpoint and Store ID.

3. Export the API key in your shell:

   export API_KEY='<API_KEY>'
   

Keep the API key out of source control, application logs, and other unsecured locations.

Install the SDK

python -m pip install redis-agent-memory

Create the client and check the service health

Create quickstart.py with the following code. Replace <ENDPOINT> and <STORE_ID> with the values from Redis Cloud. The endpoint must include https://.

import os
from datetime import datetime, timezone

from redis_agent_memory import AgentMemory, models


ENDPOINT = "<ENDPOINT>"
STORE_ID = "<STORE_ID>"
SESSION_ID = "travel-planning-session"
USER_ID = "quickstart-user"


def show(label, response):
    print(f"{label}:")
    print(response.model_dump_json(by_alias=True, indent=2))


def main():
    with AgentMemory(
        ENDPOINT,
        store_id=STORE_ID,
        api_key=os.environ["API_KEY"],
    ) as agent_memory:
        health = agent_memory.health()
        show("Service health", health)


if __name__ == "__main__":
    main()

Run the file:

python quickstart.py

A healthy response confirms that the client can reach Redis Agent Memory and authenticate with the API key. The first store request validates the Store ID.

1. Build conversation context with session memory

Session memory stores a conversation as an ordered sequence of events. Add the following code after the health check, inside the with block:

        event = agent_memory.add_session_event(
            session_id=SESSION_ID,
            actor_id=USER_ID,
            role=models.MessageRole.USER,
            content=[models.Text(
                text=(
                    "I am visiting Tokyo and Kyoto next month. "
                    "I am vegetarian and prefer spicy food."
                ),
            )],
            created_at=datetime.now(timezone.utc),
        )
        show("Created event", event)

        session = agent_memory.get_session_memory(
            session_id=SESSION_ID,
        )
        show("Session memory", session)

Run the file again. The session response contains the stored message, its role, actor, and timestamps. An application can retrieve this session before the next agent turn and add the events to the model's context.

After the event is stored, comment out the call to add_session_event before subsequent runs to avoid adding the same message again.

2. Recall automatically extracted information

Redis Agent Memory processes session events in the background and creates long term memories for information that may be useful in later conversations. You configured the extraction cadence to one minute when you created the service. You do not need to call a memory creation method.

Wait at least one minute, then add this search after the session retrieval:

        results = agent_memory.search_long_term_memory(
            request={
                "text": "What dietary requirements and food preferences does the user have?",
                "filter_": {
                    "owner_id": {
                        "eq": USER_ID,
                    }
                },
                "limit": 5,
            },
        )
        show("Automatically extracted memories", results)

Run the file. The items array should contain memories derived from the conversation, such as the vegetarian requirement or preference for spicy food. Extraction is asynchronous, so run the search again if the array is empty.

The extracted memory remains searchable after the session expires, subject to the long term memory TTL. You can change the extraction cadence and both TTLs in the Redis Agent Memory service configuration.

The Python SDK uses snake case for method arguments and request fields. Serialized API requests and responses use camel case.

3. Keep long conversations concise with automatic summarization

Automatic summarization condenses older events and retains the most recent events in full. The retrieved session then contains a summary object and the recent events array, so the application can provide useful history without filling the model's context window with every original message.

You enabled automatic summarization when you created the service. When the session reaches six events, Redis Agent Memory summarizes the older events and retains the two most recent events in full.

Add conversation turns

Add this code after the first session event to continue the conversation past the configured threshold:

        turns = [
            (models.MessageRole.ASSISTANT, "What dates are you traveling?"),
            (models.MessageRole.USER, "I arrive on October 10 and leave on October 18."),
            (models.MessageRole.ASSISTANT, "Would you like formal or casual restaurants?"),
            (models.MessageRole.USER, "Mostly casual places near public transit."),
            (models.MessageRole.ASSISTANT, "Do you have a preferred budget?"),
            (models.MessageRole.USER, "About 40 euros per person."),
        ]

        for role, text in turns:
            agent_memory.add_session_event(
                session_id=SESSION_ID,
                actor_id=USER_ID if role == models.MessageRole.USER else "travel-agent",
                role=role,
                content=[models.Text(text=text)],
                created_at=datetime.now(timezone.utc),
            )

Run the code once, then comment out the loop to avoid adding the same turns again. Summarization runs in the background.

Retrieve the summarized session

After a short wait, retrieve the session again:

        compacted_session = agent_memory.get_session_memory(
            session_id=SESSION_ID,
        )
        show("Compacted session memory", compacted_session)

Run the retrieval again after a short wait if summary is not present. Compare summary.text with the recent events. The summary should preserve earlier trip decisions while recent turns remain available in full.

See automatic summarization configuration for details.

4. Extract business specific data with a custom memory type

Built in memories preserve generally useful information. Custom memory types let an application extract structured information for its business domain. You configured trip_preference when you created the service, so it processed the same travel planning event independently.

Search for the structured memory:

        custom_results = agent_memory.search_long_term_memory(
            request={
                "text": "What are the requirements for the user's trip?",
                "filter_": {
                    "owner_id": {"eq": USER_ID},
                    "memory_type": {"eq": "trip_preference"},
                },
                "limit": 5,
            },
        )
        show("Trip preference memories", custom_results)

The result uses trip_preference as its memoryType and contains travel information extracted from the conversation. The exact text and returned fields depend on the conversation, extraction model, and client.

See custom memory types for configuration requirements and limits.

5. Guide extraction away from sensitive data

The semantic exclusion prompt tells Redis Agent Memory which information should not be kept in long-term memory. Add an event containing a fictional booking code and information that is safe to retain:

        sensitive_event = agent_memory.add_session_event(
            session_id=SESSION_ID,
            actor_id=USER_ID,
            role=models.MessageRole.USER,
            content=[models.Text(
                text=(
                    "I booked Hotel Sakura in Tokyo. For this example, "
                    "the fictional booking confirmation code is DEMO-7QX9."
                ),
            )],
            created_at=datetime.now(timezone.utc),
        )
        show("Event with excluded information", sensitive_event)

Run the code once, then comment out the call to add_session_event. Wait at least one minute and search for the safe hotel information:

        exclusion_results = agent_memory.search_long_term_memory(
            request={
                "text": "Where is the user staying in Tokyo?",
                "filter_": {
                    "owner_id": {"eq": USER_ID},
                },
                "limit": 5,
            },
        )
        show("Memories after semantic exclusion", exclusion_results)

Inspect the returned memories. They can retain the hotel name, but should not contain DEMO-7QX9 because the exclusion prompt covers booking confirmation codes.

See sensitive-data exclusions for configuration details.

Next steps

• Review the Python SDK package and reference.
• Try the TypeScript SDK quickstart or REST API quickstart.
• Learn when to create long term memories directly.