Changelog

SDK Method Name Changes

In an effort to keep our SDK method names consistent with our conventions, we have renamed the following methods:

Before and After

New Projects Endpoint

A new Projects endpoint has been added to the API, allowing you to manage projects and their associated templates.

The new endpoints can be found here: Projects

New Batch message creation API

A series of new Batch endpoints has been introduced to support batch message creation, allowing you to perform multiple LLM requests in a single API call. These APIs leverage provider batch APIs under the hood, which can be more cost-effective than making multiple API calls.

New endpoints can be found here: Batch Messages

New List Agent Groups API added

The List Agent Groups API has been added to the Agents endpoint, allowing you to retrieve all multi-agent groups associated with a specific agent.

1
from letta_client import Letta
2
client = Letta(
3
    token="YOUR_API_KEY",
4
)
5
agent_groups = client.agents.list_agent_groups(
6
    agent_id="AGENT_ID",
7
)

New reasoning_effort field added to LLMConfig

The reasoning_effort field has been added to the LLMConfig object to control the amount of reasoning the model should perform, to support OpenAI’s o1 and o3 reasoning models.

New sender_id parameter added to Message model

The Message object now includes a sender_id field, which is the ID of the sender of the message, which can be either an identity ID or an agent ID. The sender_id is expected to be passed in at message creation time.

1
from letta_client import Letta
2
client = Letta(
3
    token="YOUR_API_KEY",
4
)
5
messages = client.agents.messages.create(
6
    agent_id="AGENT_ID",
7
    messages=[
8
        MessageCreate(
9
            role="user",
10
            content="Hello, how are you?",
11
            sender_id="IDENTITY_ID",
12
        )
13
    ]
14
)

New Upsert Properties API for Identities

The Upsert Properties API has been added to the Identities endpoint, allowing you to update or create properties for an identity.

1
from letta_client import IdentityProperty, Letta
2
client = Letta(
3
    token="YOUR_TOKEN",
4
)
5
client.identities.upsert_properties(
6
    identity_id="IDENTITY_ID",
7
    request=[
8
        IdentityProperty(
9
            key="name",
10
            value="Caren",
11
            type="string",
12
        ),
13
        IdentityProperty(
14
            key="email",
15
            value="caren@example.com",
16
            type="string",
17
        )
18
    ],
19
)

New Parent Tool Rule

A new tool rule has been introduced for configuring a parent tool rule, which only allows a target tool to be called after a parent tool has been run.

1
from letta_client import Letta
2
client = Letta(
3
    token="YOUR_API_KEY",
4
)
5
agent = client.agents.create(
6
    model="openai/gpt-4o-mini",
7
    embedding="openai/text-embedding-ada-002",
8
    tool_rules=[
9
        ParentToolRule(
10
            tool_name="parent_tool",
11
            children=["child_tool"]
12
        )
13
    ]
14
)

Runs API can now be filtered by Agent ID

The Runs API now supports filtering by agent_id to retrieve all runs and all active runs associated with a specific agent.

1
from letta_client import Letta
2
client = Letta(
3
    token="YOUR_API_KEY",
4
)
5
runs = client.runs.list_active_runs(
6
    agent_id="AGENT_ID",
7
)

Add new otid field to Message API

The Message object returned by our Messages endpoints now includes an offline threading id field, a unique identifier set at creation time, which can be used by the client to deduplicate messages.

Before:

1
from letta_client import Letta, MessageCreate
2
import uuid
3
client = Letta(
4
    token="YOUR_API_KEY",
5
)
6
messages = client.agents.messages.create(
7
    agent_id="AGENT_ID",
8
    messages=[
9
        MessageCreate(
10
            role="user",
11
            content="Hello, how are you?"
12
            otid=uuid.uuid4(),
13
        )
14
    ]
15
)

New strip_messages field in Import Agent API

The Import Agent API now supports a new strip_messages field to remove messages from the agent’s conversation history when importing a serialized agent file.

1
from letta_client import Letta
2
client = Letta(
3
    token="YOUR_API_KEY",
4
)
5
client.agents.import_agent_serialized(
6
    file=open("/path/to/agent/file.af", "rb"),
7
    strip_messages=True,
8
)