create_cli_agent

Parameters

Name	Type	Description
`model`*	`str \| BaseChatModel`	LLM model to use (e.g., `'provider:model'`)
`assistant_id`*	`str`	Agent identifier for memory/state storage
`tools`	`Sequence[BaseTool \| Callable \| dict[str, Any]] \| None`	Default:`None`
`mcp_tools`	`Sequence[BaseTool] \| None`	Default:`None`
`sandbox`	`SandboxBackendProtocol \| None`	Default:`None`
`sandbox_type`	`str \| None`	Default:`None`
`system_prompt`	`str \| None`	Default:`None`
`interactive`	`bool`	Default:`True`
`auto_approve`	`bool`	Default:`False`
`auto_mode_enabled`	`bool`	Default:`False`
`interrupt_shell_only`	`bool`	Default:`False`
`shell_allow_list`	`list[str] \| None`	Default:`None`
`fs_tools`	`list[FsToolName] \| None`	Default:`None`
`enable_ask_user`	`bool`	Default:`True`
`enable_memory`	`bool`	Default:`True`
`memory_auto_save`	`bool`	Default:`True`
`enable_skills`	`bool`	Default:`True`
`enable_shell`	`bool`	Default:`True`
`enable_interpreter`	`bool`	Default:`False`
`rubric_model`	`str \| BaseChatModel \| None`	Default:`None`
`rubric_max_iterations`	`int \| None`	Default:`None`
`auto_classifier_model`	`str \| BaseChatModel \| None`	Default:`None`
`recursion_limit`	`int \| None`	Default:`None`
`checkpointer`	`BaseCheckpointSaver \| None`	Default:`None`
`mcp_server_info`	`list[MCPServerInfo] \| None`	Default:`None`
`cwd`	`str \| Path \| None`	Default:`None`
`project_context`	`ProjectContext \| None`	Default:`None`
`async_subagents`	`list[AsyncSubAgent] \| None`	Default:`None`
`goal_criteria_tools`	`Sequence[BaseTool \| Callable[..., Any]] \| None`	Default:`None`
`rubric_grader_tools`	`Sequence[BaseTool \| Callable[..., Any]] \| None`	Default:`None`

Wire CodeInterpreterMiddleware from langchain-quickjs into the main agent.

Local-mode only — passing a non-None sandbox while enable_interpreter=True raises ValueError. Subagents do not receive the interpreter in v1.

PTC (tools.* host bridge) calls bypass interrupt_on/HITL approval, so settings.interpreter_ptc is the only effective control over which host tools can be invoked from inside the REPL. js_eval itself is intentionally not gated by HITL — per-call approval would be unusably noisy and would not block PTC fan-out anyway. The "safe" preset is therefore restricted to tools that are already non-HITL outside the REPL (read-only file inspection); exposing HITL-gated tools — network fetch, subagent dispatch, shell, file writes — requires an explicit list or interpreter_ptc="all" with interpreter_ptc_acknowledge_unsafe=True.

Requires the core langchain-quickjs dependency.

LangChain Assistant

Menu

Parameters