Prompt management

In production, system prompts often need versioning and dynamic delivery. Prompt management lets you template and hot-update the prompt at runtime without changing code or restarting the service. Before handling each user request, VeADK calls the prompt manager’s get_prompt method to fetch the latest content and uses it as the agent’s system prompt.

Using CozeLoop

VeADK ships CozeloopPromptManager, which connects the system prompt to CozeLoop’s cloud prompt management. Install the CozeLoop Python SDK:

pip install cozeloop

Provide credentials via environment variables rather than hardcoding them:

export COZELOOP_WORKSPACE_ID="<your Workspace ID>"
export COZELOOP_TOKEN="<your CozeLoop token>"

Create the manager and pass it to the agent:

import os
from veadk import Agent
from veadk.prompts.prompt_manager import CozeloopPromptManager

prompt_manager = CozeloopPromptManager(
    cozeloop_workspace_id=os.environ["COZELOOP_WORKSPACE_ID"],
    cozeloop_token=os.environ["COZELOOP_TOKEN"],
    prompt_key="your_prompt_key",  # the Prompt Key created in CozeLoop
    label="production",            # the Prompt label created in CozeLoop
)

agent = Agent(name="assistant", prompt_manager=prompt_manager)

CozeLoop caches prompts locally with a one-minute refresh interval. If fetching fails, it falls back to VeADK’s default system prompt.

Custom prompt manager

To load prompts from your own configuration service, database, or files, subclass BasePromptManager and implement get_prompt. It is called before handling each user request, and the returned string becomes the agent’s system prompt.

Last modified on September 19, 2026

Overview

Get started

Core components

Observability

Extensions

Frontend

Security

Using CozeLoop

Custom prompt manager

​Using CozeLoop

​Custom prompt manager

Using CozeLoop

Custom prompt manager