Code interpretation

Optional extra for code_interpreter:

pip install --extra-index-url https://elsai-agents.elsai.ai/root/ "elsai-agents-tools[agent-core-code-interpreter]==0.3.0"

Tools

python_repl

Execute Python in a persistent REPL session where variables and imports carry over between calls

Gate code with execution_permissions on AgentConfig (tools=("python_repl",)).

Notes: Not supported on Windows (fcntl unavailable). On A2A servers, PER_CONTEXT — see Stateful tools on A2A.

Environment variables

Variable	Required	Default	Description
PYTHON_REPL_PERSISTENCE_DIR	optional	./repl_state/	Where standalone (non-A2A) REPL state is saved
PYTHON_REPL_BINARY_MAX_LEN	optional	100	Max length shown for binary output

Also respects BYPASS_TOOL_CONSENT (see Tool consent).

On A2A servers, REPL state is under $ELSAI_A2A_SESSION_DIR/<context_id>/repl/.

code_interpreter

Run Python or other languages in an isolated Bedrock AgentCore sandbox with no local machine access

Extra: agent-core-code-interpreter

Notes: On A2A servers, PER_CONTEXT via AgentCoreCodeInterpreter.

Environment variables

No dedicated environment variables.

Configured via AgentCoreCodeInterpreter(...) constructor args (for example region), not env vars.

Examples

python_repl

Requires OPENAI_API_KEY. For demos set BYPASS_TOOL_CONSENT=true (see Tool consent). Not supported on Windows.

pip install --extra-index-url https://elsai-agents.elsai.ai/root/ elsai-agents==0.3.1
pip install --extra-index-url https://core-packages.elsai.ai/root/ "elsai-model[openai]==2.1.0"
pip install --extra-index-url https://elsai-agents.elsai.ai/root/ elsai-agents-tools==0.3.0

import os

from elsai import Agent
from elsai.agent import AgentConfig
from elsai_model.openai import OpenAIModel
from elsai_tools.python_repl import python_repl

agent = Agent(
    model=OpenAIModel(
        model_id="gpt-4o-mini",
        client_args={"api_key": os.environ["OPENAI_API_KEY"]},
    ),
    tools=[python_repl],
    system_prompt=(
        "Use python_repl to run Python code. "
        "After using tools, answer the user clearly in plain text."
    ),
    config=AgentConfig(name="repl_assistant"),
)

result = agent("Use the REPL to compute the sum of squares from 1 to 10, then tell me the result.")
print(result)

code_interpreter

Requires AWS credentials and the AgentCore extra. The agent model below uses OpenAI; the sandbox runs on Bedrock AgentCore.

pip install --extra-index-url https://elsai-agents.elsai.ai/root/ elsai-agents==0.3.1
pip install --extra-index-url https://core-packages.elsai.ai/root/ "elsai-model[openai]==2.1.0"
pip install --extra-index-url https://elsai-agents.elsai.ai/root/ "elsai-agents-tools[agent-core-code-interpreter]==0.3.0"

import os

from elsai import Agent
from elsai.agent import AgentConfig
from elsai_model.openai import OpenAIModel
from elsai_tools.code_interpreter import AgentCoreCodeInterpreter

interpreter = AgentCoreCodeInterpreter(region="us-west-2")

agent = Agent(
    model=OpenAIModel(
        model_id="gpt-4o-mini",
        client_args={"api_key": os.environ["OPENAI_API_KEY"]},
    ),
    tools=[interpreter.code_interpreter],
    system_prompt=(
        "Use code_interpreter to run code in the sandbox. "
        "After using tools, answer the user clearly in plain text."
    ),
    config=AgentConfig(name="sandbox_assistant"),
)

result = agent("In the sandbox, print 'Hello, World!' and return the output.")
print(result)

Parameter reference: Code interpretation API.

See also

• Prebuilt Tools overview