CAMEL-AI

CortexDB provides persistent long-term memory for CAMEL-AI agents, enabling multi-agent conversations and role-playing sessions to retain context across interactions.

Installation

pip install cortexdb-camel

Quick Start

from cortexdb import Cortex
from cortexdb_camel import CortexDBMemory, get_cortexdb_toolkit

client = Cortex(base_url="http://localhost:3141")

# As a memory backend
memory = CortexDBMemory(client=client, tenant_id="my-session")

# As agent tools
tools = get_cortexdb_toolkit(client, tenant_id="my-session")

As a Memory Backend

Use CortexDBMemory as CAMEL's memory provider to give agents persistent recall across conversations:

from cortexdb import Cortex
from cortexdb_camel import CortexDBMemory

client = Cortex(base_url="http://localhost:3141")
memory = CortexDBMemory(
    client=client,
    tenant_id="my-session",
    namespace="roleplay",
    top_k=10,
)

# Write memory records
memory.write_record("The user prefers concise technical answers.")
memory.write_record("Project deadline is next Friday.", metadata={"priority": "high"})

# Retrieve relevant context as structured records
records = memory.get_context("project deadlines")

# Retrieve as formatted text for prompt injection
context_text = memory.get_context_text("user preferences", max_tokens=4096)

# Clear all memories
memory.clear()

Per-Agent Memory

# Each agent gets its own memory namespace
assistant_memory = CortexDBMemory(
    client=client, tenant_id="session-1", namespace="assistant",
)
user_proxy_memory = CortexDBMemory(
    client=client, tenant_id="session-1", namespace="user-proxy",
)

# Or share memory across agents
shared_memory = CortexDBMemory(
    client=client, tenant_id="session-1", namespace="shared",
)

As Agent Tools

Use get_cortexdb_toolkit to create CAMEL FunctionTool instances that agents can invoke:

from camel.agents import ChatAgent
from cortexdb import Cortex
from cortexdb_camel import get_cortexdb_toolkit

client = Cortex(base_url="http://localhost:3141")
tools = get_cortexdb_toolkit(client, tenant_id="my-session", namespace="tools")

agent = ChatAgent(
    system_message="You are a helpful assistant with long-term memory.",
    tools=tools,
)

The toolkit provides three function tools:

• cortexdb_search — Search for relevant memories using natural language
• cortexdb_store — Save important information for future retrieval
• cortexdb_forget — Remove memories for privacy or data correction

Standalone Functions

You can also use the underlying functions directly:

from cortexdb import Cortex
from cortexdb_camel import cortexdb_search, cortexdb_store, cortexdb_forget

client = Cortex(base_url="http://localhost:3141")

# Store
cortexdb_store(client, "The API key rotates every 90 days.", tenant_id="ops")

# Search
results = cortexdb_search(client, "API key rotation", tenant_id="ops")

# Forget
cortexdb_forget(client, "old API key policy", "Policy updated", tenant_id="ops")

Configuration

| Parameter | Default | Description | |---|---|---| | client | Required | Initialized Cortex client instance | | tenant_id | "default" | Tenant identifier for multi-tenant isolation | | namespace | None | Namespace for organizing memories | | top_k | 5 | Maximum results per context retrieval (memory only) |

Complete Example

from camel.agents import ChatAgent
from camel.messages import BaseMessage
from cortexdb import Cortex
from cortexdb_camel import CortexDBMemory, get_cortexdb_toolkit

client = Cortex(base_url="http://localhost:3141")

# Memory backend for automatic context
memory = CortexDBMemory(
    client=client,
    tenant_id="customer-support",
    namespace="tickets",
)

# Tools for explicit memory management
tools = get_cortexdb_toolkit(
    client=client,
    tenant_id="customer-support",
    namespace="tickets",
)

# Build a support agent with persistent memory
agent = ChatAgent(
    system_message=(
        "You are a customer support agent. Use your memory tools to store "
        "ticket details and search for similar past issues. Always check "
        "memory before answering a question."
    ),
    tools=tools,
)

# First interaction — store a new ticket
memory.write_record(
    "Ticket #1234: User reports login timeout after password reset. "
    "Resolved by clearing session cache.",
    metadata={"ticket_id": "1234", "status": "resolved"},
)

# Later interaction — the agent can search for similar issues
context = memory.get_context("login timeout after password reset")
print(f"Found {len(context)} related records")

# Agent can also use tools in conversation
response = agent.step(
    BaseMessage.make_user_message(
        role_name="User",
        content="A customer is reporting login timeouts. Have we seen this before?",
    )
)