CAMEL-AI

NiadraMemory(conversation) is an AgentMemory around the agent’s own chat memory; memory.attach(agent) makes it the agent’s memory, and memory.tools (or NiadraToolkit(conversation).get_tools()) gives the kit as FunctionTools. Python only.

Install

pip install 'niadra[camel]'   # camel-ai 0.2.90 or newer, below 0.3

The five primitives

Primitive	How the adapter wires it
Context	`get_context()` returns the pack (after the agent’s own notes, with `agent_memory=`) right after the system message, then the chat history, then the `turn_block`. Only the history is stored; the Niadra messages are made fresh on every read
Turns	Every record the agent writes is recorded: the user’s message as the customer’s turn and the assistant’s text as the agent’s; tool calls and results are not turns
Tools	`memory.tools` and `NiadraToolkit(conversation).get_tools()`: CAMEL `FunctionTool`s whose OpenAI schema is the kit’s definition, word for word, bound to the customer; the toolkit serves a workforce or a role-playing session
Verification	`conversation.verify()` before `attach()`
Handoff	`transferred_to_agent()` and `transferred_to_human()` record a transfer, for a workforce that hands the conversation over or an escalation to a person

Minimal example

"""A CAMEL ChatAgent with the customer's memory as its agent memory and the history tools."""

from camel.agents import ChatAgent
from camel.models import ModelFactory
from camel.types import ModelPlatformType, ModelType

from niadra import Niadra, phone
from niadra.integrations.camel import NiadraMemory

niadra = Niadra(channel="chat")
memory = NiadraMemory(niadra.conversation("thread-81", subject=phone("+5511912345678")))
model = ModelFactory.create(model_platform=ModelPlatformType.OPENAI, model_type=ModelType.GPT_4_1)
agent = memory.attach(ChatAgent(system_message="You are Acme's agent.", model=model, tools=memory.tools))
print(agent.step("Where is my replacement lid?").msgs[0].content)
niadra.close()

The same code is in examples/camel_agent.py.

Agent memory

NiadraMemory(conversation, agent_memory=True) puts the agent’s own notes before the customer’s context and adds the two tools. See Agent memory.

Limits

CAMEL reads an agent’s memory synchronously, even in astep: the context comes with a conversation of the sync client (Niadra). With AsyncNiadra, the agent gets the tools and the turns and runs without the pack. The read has the SDK’s own time budget: 300 ms for chat and 150 ms for voice.
The camel extra does not install together with google-genai, google-adk, pipecat, livekit and openai-agents: camel-ai pins pydantic to 2.12.0 or older, and those extras want a newer one. Use separate environments.
Nothing here stops the agent: with Niadra slow or down, it runs without the pack and a tool answers that the history is unavailable.
Tested against camel-ai 0.2.90 with the model replaced by a fake and Niadra on the emulator.

Get started

Voice

Channels

Agent frameworks

Model SDKs

Automation platforms

MCP clients

Install

The five primitives

Minimal example

Agent memory

Limits

Next steps

History navigation

Agent memory

​Install

​The five primitives

​Minimal example

​Agent memory

​Limits

​Next steps

History navigation

Agent memory

Install

The five primitives

Minimal example

Agent memory

Limits

Next steps