Integrate Capsule: lightweight Wasm sandbox to run untrusted code

[mavdol]

Hi @MervinPraison,

I built a runtime to sandbox AI agent tasks. I think it could be a great lightweight alternative to run untrusted code locally for PraisonAI.

Here's an example using the LangChain integration (langchain-capsule):

from langchain_capsule import CapsulePythonTool

code = """
def factorial(n):
    if n <= 1:
        return 1
    return n * factorial(n - 1)

factorial(6)
"""

python_tool = CapsulePythonTool()
result = python_tool.run(code)
print(result) # "720"

Only the first run takes about a second (cold start), then every next run starts in ~10ms.

Why this could be useful for PraisonAI:

• Simple to adopt – No setup, no cloud dependency and works the same everywhere (dev, CI, prod).
• No Docker complexity – When running containers in production (e.g., via Kubernetes), using another Docker to isolate untrusted code adds complexity. It avoids Docker-in-Docker complexity entirely.
• Strong isolation for untrusted code – Each execution runs in its own WebAssembly sandbox, with memory space isolation and no host access.

I see PraisonAI is compatible with LangChain, but I'd be happy to build a custom integration specifically for PraisonAI if that's more suitable.

Here are the relevant links:

LangChain integration: github.com/mavdol/langchain-capsule
Main Capsule repo: github.com/mavdol/capsule

Hope this helps!

[MervinPraison]

@mavdol can you please create your pull request in https://github.com/MervinPraison/PraisonAI-Tools

[mavdol]

Thanks @MervinPraison! Perfect, I'll prepare and open the PR in PraisonAI-Tools