Assistants

1What are Assistants and their core values

AssistantIt is a concept provided by LangGraph Platform: it binds "Graph" and "Configuration" together, persists them in PostgreSQL, and allocates a uniqueassistant_id。

One sentence to understand:The same picture + different config = different Assistant。

• No need to deploy a new diagram for each business scenario, just create different Assistant instances
• Each Assistant is independently versioned - modifying the configuration will generate a new version, and the old version can be traced back
• Different Assistants hold their own threads (Threads) and memories (Stores), and data is naturally isolated.

2Prerequisite: Configuration design of the graph

Assistants are defined internally in the dependency graphconfigurable field. task_mAIstro picture inconfiguration.pyThree configurable fields are declared in and read within the node:

# configuration.py（简化）
from dataclasses import dataclass, field
from langgraph.config import get_configurable

@dataclass
class Configuration:
    user_id: str = "default_user"
    todo_category: str = "general"      # "personal" | "work" | ...
    task_maistro_role: str = ""          # 系统提示词中的角色说明

# 在图节点内读取
config = Configuration(**get_configurable())

3Create Assistant (create)

useclient.assistants.create()Create an Assistant, passing in the graph name and initial configuration.

from langgraph_sdk import get_client

client = get_client(url="http://localhost:8123")

# 创建 personal 助手（version 1）
personal_assistant = await client.assistants.create(
    "task_maistro",   # 已部署的图名称
    config={"configurable": {"todo_category": "personal"}}
)
print(personal_assistant["assistant_id"])   # uuid
print(personal_assistant["version"])         # 1

create()The returned object containsassistant_id、version、configand persist to PostgreSQL.

4Versioned update assistant (update)

useclient.assistants.update()Add new configuration to existing Assistant, each call will generate anew version, the old version is automatically retained.

task_maistro_role = """You are a friendly and organized personal task assistant.

- When providing a 'todo summary':
  1. List tasks grouped by deadline (overdue, today, this week, future)
  2. Highlight tasks missing deadlines
- Proactively ask for deadlines when new tasks lack them
- Tone: encouraging and supportive"""

configurations = {
    "todo_category": "personal",
    "user_id": "lance",
    "task_maistro_role": task_maistro_role,
}

personal_assistant = await client.assistants.update(
    personal_assistant["assistant_id"],   # 传入已有 ID
    config={"configurable": configurations}
)
print(personal_assistant["version"])  # 2

5Practical combat: Personal and Work assistants

Typical use case of task_mAIstro: Use two Assistants to manage personal tasks and work tasks respectively, isolated from each other.

Work Assistant example: Enforcing deadlines

user_input = "Create another ToDo: Finalize set of report generation tutorials."
thread = await client.threads.create()

async for chunk in client.runs.stream(
    thread["thread_id"],
    work_assistant_id,   # 使用 work assistant 的 ID
    input={"messages": [HumanMessage(content=user_input)]},
    stream_mode="values"
):
    if chunk.event == 'values':
        convert_to_messages(chunk.data["messages"])[-1].pretty_print()

# 用户在同一线程回复截止日期
user_input = "OK, for this task let's get it done by next Tuesday."
async for chunk in client.runs.stream(
    thread["thread_id"],   # 复用同一线程，保持上下文
    work_assistant_id,
    input={"messages": [HumanMessage(content=user_input)]},
    stream_mode="values"
):
    if chunk.event == 'values':
        convert_to_messages(chunk.data["messages"])[-1].pretty_print()

Personal Assistant Example: Encouraging ToDo Summary

user_input = "Give me a todo summary."
thread = await client.threads.create()
async for chunk in client.runs.stream(
    thread["thread_id"],
    personal_assistant_id,   # 切换到 personal assistant
    input={"messages": [HumanMessage(content=user_input)]},
    stream_mode="values"
):
    if chunk.event == 'values':
        convert_to_messages(chunk.data["messages"])[-1].pretty_print()

6Search and Manage Assistants

All Assistants are persisted in PostgreSQL and can be queried and deleted at any time through the SDK.

Search all Assistants

assistants = await client.assistants.search()
for assistant in assistants:
    print({
        'assistant_id': assistant['assistant_id'],
        'version': assistant['version'],
        'config': assistant['config']
    })

Delete Assistant

# 先创建一个临时 assistant
temp_assistant = await client.assistants.create(
    "task_maistro",
    config={"configurable": configurations}
)

# 删除它
await client.assistants.delete(temp_assistant["assistant_id"])

# 确认已删除
assistants = await client.assistants.search()
print(len(assistants))  # 回到删除前的数量

Get assistant_id through search

# 按返回顺序分配（通常按创建时间倒序）
work_assistant_id     = assistants[0]['assistant_id']
personal_assistant_id = assistants[1]['assistant_id']

In a production environment, it is recommended toassistant_idPersistence to the application database instead of retrieving fromsearch()Inferred indexing - ordering may be unstable when multiple people are collaborating.

7Architecture summary and applicable scenarios