Camel AI integration guide

​

Prerequisites

• A SambaCloud API key
  Get one by creating a SambaCloud account and generating an API key from the API Keys tab.
• Python 3.10 to 3.14. camel-ai does not support Python 3.9 or earlier, or Python 3.15 or later.
• Your API key exported as SAMBA_API_KEY:
  export SAMBA_API_KEY="your-api-key"
  

​

Installation and setup

1. Create a virtual environment
   python3 -m venv .venv
   source .venv/bin/activate
   
2. Confirm the environment uses a supported version
   python --version
   
3. Install the required libraries. Quote "camel-ai[all]~=0.2.90": zsh, the default shell on macOS, treats the square brackets as a filename pattern and fails with no matches found before pip runs. The extra pulls a large dependency tree and can take several minutes, so allow time for it to finish.
   pip install "camel-ai[all]~=0.2.90" "mcp<2" starlette nest_asyncio jupyter ipywidgets
   
   Both version bounds are deliberate. Keep them.~=0.2.90 holds you on the camel-ai 0.2 line. camel-ai is pre-1.0, so its public API can change between releases. The code on this page is verified against camel-ai 0.2.90. Widen the bound only after you have re-tested your own code."mcp<2" is required. camel-ai requests mcp>=1.3.0 with no upper bound, so an unpinned install pulls mcp 2.0.0, which removed FastMCP from mcp.server. Without the pin, from camel.agents import ChatAgent fails with ImportError: cannot import name 'FastMCP' from 'mcp.server', and neither the examples below nor the notebook can run. Remove the pin once camel-ai bounds the dependency itself.
4. Verify the installation. This imports ChatAgent rather than the top-level package, because import camel succeeds even when the agent API is broken.
   python -c "from camel.agents import ChatAgent; print('Camel-AI ready')"
   
   The command prints exactly this and exits without a traceback:
   Camel-AI ready
   

​

Quickstart

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

model = ModelFactory.create(
    model_platform=ModelPlatformType.SAMBA,
    model_type="Meta-Llama-3.3-70B-Instruct",
    url="https://api.sambanova.ai/v1",
)
agent = ChatAgent(model=model)
print(agent.step("Say hello in five words.").msgs[0].content)

Hello, how are you today?

​

Coordinate two agents

import os
import sys

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

if not os.environ.get("SAMBA_API_KEY"):
    sys.exit("Set SAMBA_API_KEY before running this script.")

model = ModelFactory.create(
    model_platform=ModelPlatformType.SAMBA,
    model_type="Meta-Llama-3.3-70B-Instruct",
    url="https://api.sambanova.ai/v1",
)

planner = ChatAgent(
    system_message=(
        "You are a release planner. Given a goal, reply with a numbered plan "
        "of at most four steps. Keep each step to one sentence."
    ),
    model=model,
)
reviewer = ChatAgent(
    system_message=(
        "You are a critical reviewer. Given a plan, name the single riskiest "
        "step and propose one concrete change. Reply in under 80 words."
    ),
    model=model,
)

goal = "Ship a rate limiter for a public REST API."

try:
    plan = planner.step(goal).msgs[0].content
    critique = reviewer.step(plan).msgs[0].content
    revised = planner.step(
        f"Revise your plan using this feedback:\n{critique}"
    ).msgs[0].content
except Exception as exc:
    sys.exit(f"CAMEL request failed: {exc}")

print("PLAN\n", plan)
print("\nREVIEW\n", critique)
print("\nREVISED PLAN\n", revised)

PLAN
 1. Define the rate limit policy and the window granularity.
 ...

REVIEW
 The riskiest step is ... Instead, ...

REVISED PLAN
 1. Define the rate limit policy per API key rather than per IP.
 ...

​

Example notebook

• Customer Service Discord Bot with Agentic RAG

• A Firecrawl account for FIRECRAWL_API_KEY, used to crawl the knowledge source.
• A bot registered in the Discord Developer Portal for DISCORD_BOT_TOKEN.
• A GPU runtime. The notebook runs an open-source embedding model locally and suggests a GPU-backed Google Colab runtime (Tesla T4) for that step. You can substitute another embedding model supported by CAMEL-AI instead.

​

Troubleshooting

​

Additional resources

• CAMEL-AI documentation for the full framework reference.
• CAMEL-AI societies for the RolePlaying and Workforce patterns that go beyond the two-agent example above.
• camel-ai on GitHub to check the source for a signature or file an upstream issue.
• SambaCloud models for current model IDs and context lengths.