langchain-agentcore-codeinterpreter
Amazon Bedrock AgentCore Code Interpreter sandbox integration for Deep Agents.
This package provides AgentCoreSandbox — a SandboxBackendProtocol implementation that wraps AgentCore's Code Interpreter, a secure, isolated MicroVM environment for executing code. AgentCoreSandbox.create() starts a session and stops it when you close the sandbox or leave a with block.
Note: For the LangChain
BaseToolintegration (used withcreate_react_agentand LangGraph agents), seelangchain-aws[tools]. This package is specifically for the Deep AgentsBaseSandboxprotocol.
Prerequisites
1. AWS credentials configured via one of the following methods:
# Option 1: Long-lived IAM credentials
export AWS_ACCESS_KEY_ID="your-access-key"
export AWS_SECRET_ACCESS_KEY="your-secret-key"
export AWS_REGION="us-west-2"
# Option 2: Temporary credentials (IAM roles, SSO, STS AssumeRole)
export AWS_ACCESS_KEY_ID="your-access-key"
export AWS_SECRET_ACCESS_KEY="your-secret-key"
export AWS_SESSION_TOKEN="your-session-token"
export AWS_REGION="us-west-2"
# Option 3: AWS CLI profile (picks up ~/.aws/credentials + ~/.aws/config)
aws configure
# or for SSO:
aws configure sso
aws sso login --profile your-profile
Any method supported by the boto3 credential chain works, including EC2 instance profiles, ECS task roles, and environment variables.
2. IAM permissions — your credentials must allow bedrock-agentcore:InvokeCodeInterpreter (or the equivalent action for your region). See the AgentCore Code Interpreter docs for the required IAM policy.
3. Region availability — Code Interpreter is available in select AWS regions. us-west-2 is a safe default. Pass the region to AgentCoreSandbox.create(region=...).
Quick Install
pip install langchain-agentcore-codeinterpreter
Usage
Standalone
from langchain_agentcore_codeinterpreter import AgentCoreSandbox
with AgentCoreSandbox.create(region="us-west-2") as backend:
result = backend.execute("echo hello")
print(result.output, result.exit_code) # "hello" 0
Sessions last 15 minutes by default. Pass session_timeout_seconds (up to 28,800, or 8 hours) for longer agent runs, and identifier to use a custom code interpreter with VPC networking or an execution role.
To configure the CodeInterpreter client yourself, pass a started one to the constructor instead. You then own its lifecycle:
from bedrock_agentcore.tools.code_interpreter_client import CodeInterpreter
interpreter = CodeInterpreter(
region="us-west-2", integration_source="langchain-deepagents"
)
interpreter.start()
try:
backend = AgentCoreSandbox(interpreter=interpreter)
...
finally:
interpreter.stop()
With Deep Agents
from deepagents import create_deep_agent
from langchain_agentcore_codeinterpreter import AgentCoreSandbox
from langchain_aws import ChatBedrockConverse
model = ChatBedrockConverse(
model="us.anthropic.claude-sonnet-4-6",
region_name="us-west-2",
)
with AgentCoreSandbox.create(
region="us-west-2", session_timeout_seconds=3600
) as backend:
agent = create_deep_agent(
model=model,
backend=backend,
system_prompt="You are a coding assistant with sandbox access.",
)
result = agent.invoke(
{
"messages": [
{"role": "user", "content": "Create and run a hello world script"}
]
}
)
print(result["messages"][-1].content)
File operations
# Upload files
backend.upload_files(
[
("data.csv", b"name,value\nalice,42\nbob,17"),
("analyze.py", b"import csv\nprint('ready')"),
]
)
# Download files
results = backend.download_files(["data.csv"])
for r in results:
if r.content is not None:
print(f"{r.path}: {r.content.decode()}")
else:
print(f"Failed: {r.path}: {r.error}")
Session behavior
AgentCore sessions cannot be reconnected once stopped. Each session is a fresh, isolated MicroVM, and its files are gone when it ends. Sessions auto-expire after session_timeout_seconds (default 15 minutes, maximum 8 hours); calls after that return a session-expired error.
execute() honors timeout by running the command under GNU timeout inside the sandbox. A command stopped this way returns exit code 124.
AgentCore caps each request and response at 100 MB and gives each session 10 GB of disk. See AgentCore quotas.
Contributing
See the langchain-aws contributing guide.
cd libs/agentcore-codeinterpreter
# Run unit tests (no network, no AWS credentials needed)
make tests
# Run linter
make lint
# Run integration tests (requires AWS credentials)
make integration_tests
Metadata
Release files for langchain-agentcore-codeinterpreter 0.0.6
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| langchain_agentcore_codeinterpreter-0.0.6.tar.gz | 220.5 kB | Without provenance | Details for langchain_agentcore_codeinterpreter-0.0.6.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| langchain_agentcore_codeinterpreter-0.0.6-py3-none-any.whl | Python 3 | none | any | Details for langchain_agentcore_codeinterpreter-0.0.6-py3-none-any.whl |
Total release size: 234.2 kB
Release files / langchain_agentcore_codeinterpreter-0.0.6.tar.gz
| Download URL | langchain_agentcore_codeinterpreter-0.0.6.tar.gz |
|---|---|
| Size | 220.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
8ba25fa3669bf09551fbabcfd40d0e2888808a96bacab1e9bb151b5b2b9ca419
|
|
BLAKE2b-256 checksum How to use checksums |
41288da31d35834769baba24f67e3631ebd431ee86dc5d6dbe669beffb3c7980
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Release files / langchain_agentcore_codeinterpreter-0.0.6-py3-none-any.whl
| Download URL | langchain_agentcore_codeinterpreter-0.0.6-py3-none-any.whl |
|---|---|
| Size | 13.7 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
3fc837c445d2dbd33510c99cedcc4759933561482cf07518c1ba2515811d6a70
|
|
BLAKE2b-256 checksum How to use checksums |
da91d4b3459fbb0104eadc79dc06b4f58ca421cb0011a12a0e1ac1bef3405f39
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Security for langchain-agentcore-codeinterpreter 0.0.6
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.