Skip to content

armgit / camel

Public Imported from GitHub
0

🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org

14.8 KB
Raw Blame

title: "Memory" icon: memory

doc_code_map: - "camel/memories/**/*.py"

The CAMEL Memory module gives your AI agents a flexible, persistent way to **store, retrieve, and manage information**, across any conversation or task.

With memory, agents can maintain context, recall key details from previous chats, and deliver much more coherent, context-aware responses.

Memory is what transforms a “chatbot” into a smart, adaptable assistant.

This is the fastest way to enable true memory for your agents: store, retrieve, and leverage context across interactions. ```python from camel.memories import ( ChatHistoryBlock, LongtermAgentMemory, MemoryRecord, ScoreBasedContextCreator, VectorDBBlock, ) from camel.messages import BaseMessage from camel.types import ModelType, OpenAIBackendRole from camel.utils import OpenAITokenCounter
  # Initialize the memory
  memory = LongtermAgentMemory(
      context_creator=ScoreBasedContextCreator(
          token_counter=OpenAITokenCounter(ModelType.GPT_4O_MINI),
          token_limit=1024,
      ),
      chat_history_block=ChatHistoryBlock(),
      vector_db_block=VectorDBBlock(),
  )

  # Create and write new records
  records = [
      MemoryRecord(
          message=BaseMessage.make_user_message(
              role_name="User",
              meta_dict=None,
              content="What is CAMEL AI?",
          ),
          role_at_backend=OpenAIBackendRole.USER,
      ),
      MemoryRecord(
          message=BaseMessage.make_assistant_message(
              role_name="Agent",
              meta_dict=None,
              content="CAMEL-AI.org is the 1st LLM multi-agent framework and "
                      "an open-source community dedicated to finding the scaling law "
                      "of agents.",
          ),
          role_at_backend=OpenAIBackendRole.ASSISTANT,
      ),
  ]
  memory.write_records(records)

  # Get context for the agent
  context, token_count = memory.get_context()

  print(context)
  print(f"Retrieved context (token count: {token_count}):")
  for message in context:
      print(f"{message}")
  ```
</Tab>
<Tab title="Output">
  ```markdown
  >>> Retrieved context (token count: 49):
  {'role': 'user', 'content': 'What is AI?'}
  {'role': 'assistant', 'content': 'AI refers to systems that mimic human intelligence.'}
  ```
</Tab>
Assign memory to any agent and watch your AI recall and reason like a pro. ```python from camel.agents import ChatAgent
  # Define system message for the agent
  sys_msg = BaseMessage.make_assistant_message(
      role_name='Agent',
      content='You are a curious agent wondering about the universe.',
  )

  # Initialize agent
  agent = ChatAgent(system_message=sys_msg)

  # Set memory to the agent
  agent.memory = memory

  # Define a user message
  usr_msg = BaseMessage.make_user_message(
      role_name='User',
      content="Tell me which is the 1st LLM multi-agent framework based on what we have discussed",
  )

  # Sending the message to the agent
  response = agent.step(usr_msg)

  # Check the response (just for illustrative purpose)
  print(response.msgs[0].content)
  ```
</Tab>
<Tab title="Output">
  ```markdown
  >>> CAMEL AI is recognized as the first LLM (Large Language Model) multi-agent framework. It is an open-source community initiative focused on exploring the scaling laws of agents, enabling the development and interaction of multiple AI agents in a collaborative environment. This framework allows researchers and developers to experiment with various configurations and interactions among agents, facilitating advancements in AI capabilities and understanding.
  ```
</Tab>
**What it is:** The basic data unit in CAMEL’s memory system—everything stored/retrieved flows through this structure.
**Attributes:**
- **message**: The content, as a `BaseMessage`
- **role_at_backend**: Backend role (`OpenAIBackendRole`)
- **uuid**: Unique identifier for the record
- **extra_info**: Optional metadata (key-value pairs)

**Key methods:**
- `from_dict()`: Build from a Python dict
- `to_dict()`: Convert to dict for saving/serialization
- `to_openai_message()`: Transform into an OpenAI message object
**What it is:** Result of memory retrieval from `AgentMemory`, scored for context relevance.
**Attributes:**
- **memory_record**: The original `MemoryRecord`
- **score**: How important/relevant this record is (float)
**What it is:** The core “building block” for agent memory, following the Composite design pattern (supports tree structures).
**Key methods:**
- `write_records()`: Store multiple records
- `write_record()`: Store a single record
- `clear()`: Remove all stored records
**What it is:** Defines strategies for generating agent context when data exceeds model limits.
**Key methods/properties:**
- `token_counter`: Counts message tokens
- `token_limit`: Max allowed tokens in the context
- `create_context()`: Algorithm for building context from chat history
**What it is:** Specialized `MemoryBlock` for direct agent use.
**Key methods:**
- `retrieve()`: Get `ContextRecord` list
- `get_context_creator()`: Return the associated context creator
- `get_context()`: Return properly sized chat context
**What it does:** Stores and retrieves recent chat history (like a conversation timeline).
**Initialization:**
- `storage`: Optional (default `InMemoryKeyValueStorage`)
- `keep_rate`: Historical message score weighting (default `0.9`)

**Methods:**
- `retrieve()`: Get recent chats (windowed)
- `write_records()`: Add new records
- `clear()`: Remove all chat history

**Use Case:**
Best for maintaining the most recent conversation flow/context.
**What it does:** Uses vector embeddings for storing and retrieving information based on semantic similarity.
**Initialization:**
- `storage`: Optional vector DB (`QdrantStorage` by default)
- `embedding`: Embedding model (default: `OpenAIEmbedding`)

**Methods:**
- `retrieve()`: Get similar records based on query/keyword
- `write_records()`: Add new records (converted to vectors)
- `clear()`: Remove all vector records

**Use Case:**
Ideal for large histories or when semantic search is needed.
**Key Differences:** - **Storage:** ChatHistoryBlock uses key-value storage. VectorDBBlock uses vector DBs. - **Retrieval:** ChatHistoryBlock retrieves by recency. VectorDBBlock retrieves by similarity. - **Data:** ChatHistoryBlock stores raw messages. VectorDBBlock stores embeddings. **What is it?** An **AgentMemory** implementation that wraps `ChatHistoryBlock`. **Best for:** Sequential, recent chat context (simple conversation memory).

Initialization:

  • context_creator: BaseContextCreator
  • storage: Optional BaseKeyValueStorage
  • window_size: Optional int (retrieval window)

Methods:

  • retrieve(): Get recent chat messages
  • write_records(): Write new records to chat history
  • get_context_creator(): Get the context creator
  • clear(): Remove all chat messages
**What is it?** An **AgentMemory** implementation that wraps `VectorDBBlock`. **Best for:** Semantic search—find relevant messages by meaning, not just recency.

Initialization:

  • context_creator: BaseContextCreator
  • storage: Optional BaseVectorStorage
  • retrieve_limit: int (default 3)

Methods:

  • retrieve(): Get relevant messages from the vector DB
  • write_records(): Write new records and update topic
  • get_context_creator(): Get the context creator
**What is it?** Combines **ChatHistoryMemory** and **VectorDBMemory** for hybrid memory. **Best for:** Production bots that need both recency & semantic search.

Initialization:

  • context_creator: BaseContextCreator
  • chat_history_block: Optional ChatHistoryBlock
  • vector_db_block: Optional VectorDBBlock
  • retrieve_limit: int (default 3)

Methods:

  • retrieve(): Get context from both history & vector DB
  • write_records(): Write to both chat history & vector DB
  • get_context_creator(): Get the context creator
  • clear(): Remove all records from both memory blocks
Add [Mem0](https://mem0.ai/) for cloud-based memory with automatic sync.

Initialization Params:

  • api_key: (optional) Mem0 API authentication
  • agent_id: (optional) Agent association
  • user_id: (optional) User association
  • metadata: (optional) Dict of metadata for all memories
```python from camel.memories import ChatHistoryMemory, ScoreBasedContextCreator from camel.storages import Mem0Storage from camel.types import ModelType from camel.utils import OpenAITokenCounter
memory = ChatHistoryMemory(
    context_creator=ScoreBasedContextCreator(
        token_counter=OpenAITokenCounter(ModelType.GPT_4O_MINI),
        token_limit=1024,
    ),
    storage=Mem0Storage(
        api_key="your_mem0_api_key",  # Or set MEM0_API_KEY env var
        agent_id="agent123"
    ),
    agent_id="agent123"
)
# ...write and retrieve as usual...
```
```markdown >>> Retrieved context (token count: 49): {'role': 'user', 'content': 'What is CAMEL AI?'} {'role': 'assistant', 'content': 'CAMEL-AI.org is the 1st LLM multi-agent framework and an open-source community dedicated to finding the scaling law of agents.'} ```

Why use this?

  • Cloud persistence of chat history
  • Simple setup and config
  • Sequential retrieval—conversation order preserved
  • Syncs across sessions automatically

Use when: you need reliable, persistent chat history in the cloud (not advanced semantic search).

[Memanto](https://docs.memanto.ai) provides semantic long-term memory through `MemantoMemory`, an implementation of CAMEL's agent memory interface. `ChatAgent` automatically archives user and assistant text and retrieves relevant memories using the current user message.

Start a Memanto server and create an agent once:

pip install memanto
memanto serve
# In another terminal:
memanto agent create my-camel-agent
from camel.agents import ChatAgent
from camel.memories import MemantoMemory, ScoreBasedContextCreator
from camel.types import ModelType
from camel.utils import OpenAITokenCounter

memory = MemantoMemory(
    context_creator=ScoreBasedContextCreator(
        OpenAITokenCounter(ModelType.GPT_4O_MINI), token_limit=4096,
    ),
    agent_id="my-camel-agent",
    base_url="http://localhost:8000",
    retrieve_limit=3,
)
try:
    agent = ChatAgent(
        system_message="You are a helpful assistant.",
        agent_id="my-camel-agent",
        memory=memory,
    )
    response = agent.step("I prefer concise Python examples.")
    print(response.msgs[0].content)
finally:
    memory.close()

Set OPENAI_API_KEY for the default ChatAgent model. agent_id and base_url can also be configured through MEMANTO_AGENT_ID and MEMANTO_BASE_URL.

Complete current conversation records, including system roles, attachments, and tool calls/results, remain in the configured chat-history storage (in-memory by default). Memanto stores user/assistant text and returns it as historical context; it is not a lossless backup of the conversation.

clear(), ChatAgent.reset(), and rollback operations affect current chat history only. Archived Memanto memories remain available for later recall. Manage deletion of archived memories in Memanto. Remote writes and retrieval can depend on server indexing latency; current user input remains available in local chat history immediately. HTTP errors are raised to the caller.

See the memory reference.

You can subclass `BaseContextCreator` for advanced control. ```python from camel.memories import BaseContextCreator
class MyCustomContextCreator(BaseContextCreator):
    @property
    def token_counter(self):
        # Implement your token counting logic
        return

    @property
    def token_limit(self):
        return 1000

    def create_context(self, records):
        # Implement your context creation logic
        pass
```
You can use custom embeddings or vector DBs. ```python from camel.embeddings import OpenAIEmbedding from camel.memories import VectorDBBlock from camel.storages import QdrantStorage
vector_db = VectorDBBlock(
    embedding=OpenAIEmbedding(),
    storage=QdrantStorage(vector_dim=OpenAIEmbedding().get_output_dim()),
)
```
- For production, use persistent storage (not just in-memory). - Optimize your context creator for both relevance and token count.