langchain-ai/deepagents · error · ValueError

enable_interpreter=True is not supported with a remote sandb

Error message

enable_interpreter=True is not supported with a remote sandbox in this release. Disable the sandbox or unset enable_interpreter.

What it means

create_cli_agent refuses to build an agent when enable_interpreter=True is combined with a remote sandbox. The code-interpreter middleware is only wired for local execution in this release, so the combination is explicitly rejected with a ValueError before any provider calls are made. This is a guard against a known-unsupported configuration, not a transient failure.

Source

Thrown at libs/code/deepagents_code/agent.py:2993

                env=shell_env,
            )
        else:
            # No shell access - use plain FilesystemBackend
            backend = FilesystemBackend(root_dir=root_dir, virtual_mode=False)
    else:
        # ========== REMOTE SANDBOX MODE ==========
        backend = sandbox  # Remote sandbox (ModalSandbox, etc.)
        # Note: Shell middleware not used in sandbox mode
        # File operations and execute tool are provided by the sandbox backend

    if enable_interpreter:
        if sandbox is not None:
            msg = (
                "enable_interpreter=True is not supported with a remote "
                "sandbox in this release. Disable the sandbox or unset "
                "enable_interpreter."
            )
            raise ValueError(msg)
        # Lazy import keeps `dcode -v` fast — see AGENTS.md startup-perf rule.
        from langchain_core._api import (  # noqa: PLC2701  # re-exported in _api.__all__
            suppress_langchain_beta_warning,
        )
        from langchain_quickjs import CodeInterpreterMiddleware, PTCOption

        interpreter = interpreter_config or InterpreterConfig.from_resolver()
        ptc_names = _resolve_ptc_option(
            interpreter.ptc,
            tools=tools,
            acknowledge_unsafe=interpreter.ptc_acknowledge_unsafe,
            auto_approve=auto_approve,
        )
        ptc_option: PTCOption | None = (
            cast("PTCOption", list(ptc_names)) if ptc_names is not None else None
        )
        # `CodeInterpreterMiddleware` is decorated `@beta()`, which emits a
        # `LangChainBetaWarning` on every instantiation. We intentionally use it

View on GitHub (pinned to a1af029e6e)

Solutions

Unset enable_interpreter (leave it False/default) while keeping the remote sandbox enabled.
Disable the sandbox (pass sandbox=None) if you need enable_interpreter=True.
Check the release notes/changelog for when remote-sandbox interpreter support lands, and upgrade before re-enabling the combination.

Example fix

// before
agent = create_cli_agent(sandbox=remote_sandbox, enable_interpreter=True)

// after (option A: keep remote sandbox)
agent = create_cli_agent(sandbox=remote_sandbox)
// after (option B: keep interpreter, local execution)
agent = create_cli_agent(enable_interpreter=True)

Defensive patterns

Strategy: validation

Validate before calling

if sandbox is not None and getattr(options, 'enable_interpreter', False):
    # reconfigure before calling create_cli_agent
    options.enable_interpreter = False  # or drop the remote sandbox

Type guard

def interpreter_compatible(sandbox: object | None, enable_interpreter: bool) -> bool:
    return not (sandbox is not None and enable_interpreter)

Prevention

Keep enable_interpreter at its default unless running fully locally.
Centralize agent construction in one factory so sandbox/interpreter flags are checked in one place.
Read the release notes before combining experimental flags (interpreter) with remote infrastructure.

When it happens

Trigger: Calling create_cli_agent (directly or via build_agent / _create_cli_graphs_sync) with sandbox=<remote sandbox object> and enable_interpreter=True set in the agent options.

Common situations: Enabling the PTC/quickjs interpreter in a config or CLI flag that also points at a hosted/remote sandbox; copying an example that sets enable_interpreter=True into an environment that already uses a remote sandbox; upgrading to a release where interpreter support for remote sandboxes has not shipped yet.

Related errors

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/9a7cc8d7d7340b39. Report an issue: GitHub.