---
title: Code Agents
slug: /bundles-codeagents
---

import Icon from "@site/src/components/icon";

<Icon name="Blocks" aria-hidden="true" /> [**Bundles**](/components-bundle-components) contain custom components that support specific third-party integrations with Langflow.

Langflow integrates with [OpenDsStar](https://github.com/IBM/OpenDsStar) through a bundle of code-writing agent components that can iteratively plan, write, and execute Python code to solve complex tasks.

:::info
Code Agents components are in **beta** and can change in future releases.
:::

## Prerequisites

* **OpenDsStar package**: Code Agents components require the `OpenDsStar` package and Python 3.11 or later.

    Install the dependency with:

    ```bash
    uv pip install OpenDsStar
    ```

    For more information, see [Install custom dependencies](/install-custom-dependencies).

## Use Code Agents components in a flow

For an example of using this component, see the **Structured Data Agent** starter template.

## Code Agents components

The following sections describe the purpose and configuration options for each component in the **Code Agents** bundle.

### CodeAct Agent (Smolagents)

The **CodeAct Agent (Smolagents)** component is a code-writing agent that uses the [smolagents](https://huggingface.co/docs/smolagents) `CodeAgent` to solve tasks by iteratively generating and executing Python code.

At each step the agent writes code, executes it in a sandboxed interpreter, and uses the output to determine the next action until it reaches a final answer.

The component outputs a [`Message`](/data-types#message) containing the final answer.

#### CodeAct Agent (Smolagents) parameters

| Name | Type | Description |
|------|------|-------------|
| input_value | String | Input parameter. The message or query to send to the agent. |
| llm | LanguageModel | Input parameter. The language model to use for code generation. Required. |
| tools | Tool | Input parameter. Optional tools the agent can call in addition to executing code. |
| max_iterations | Integer | Input parameter. Maximum number of agent iterations. Default: `5`. Range: 1–50. |
| system_prompt | String | Input parameter. A system message to customize agent behavior. |
| code_timeout | Integer | Input parameter. Maximum time in seconds for each code execution step. Default: `30`. Range: 5–300. |
| show_code_steps | String | Input parameter. How to display coding steps in the **Playground**: `All Steps`, `Final Code Only`, or `None`. Default: `All Steps`. |
| handle_parsing_errors | Boolean | Input parameter. Whether the agent should attempt to recover from input parsing errors. Default: `true`. |
| verbose | Boolean | Input parameter. Enable verbose logging. Default: `true`. |
| chat_history | Data | Input parameter. Chat history for multi-turn conversations. |
| agent_description | String | Input parameter. Description of the agent shown when used in Tool Mode. |

### OpenDsStar Agent

The **OpenDsStar Agent** component is a structured data analysis agent that uses [OpenDsStar](https://github.com/IBM/OpenDsStar) and LangGraph to answer complex questions over structured data files.

The agent executes a multi-step LangGraph workflow: planning → code generation → code execution → verification → routing → finalization.
Each step is streamed to the **Playground** as it runs.
Charts and images returned by the agent are rendered inline.

The component outputs a [`Message`](/data-types#message) containing the final answer.

#### OpenDsStar Agent parameters

| Name | Type | Description |
|------|------|-------------|
| input_value | String | Input parameter. The message or query to send to the agent. |
| llm | LanguageModel | Input parameter. The language model to use for the agent. Required. |
| tools | Tool | Input parameter. Optional tools the agent can use, such as **File Content Retriever** or vector store search. |
| max_iterations | Integer | Input parameter. Maximum number of agent steps. Default: `10`. Range: 1–100. |
| code_execution_mode | String | Input parameter. Code execution mode. `stepwise` executes each step separately; `full` executes all steps together. Default: `stepwise`. |
| system_prompt | String | Input parameter. A system message to customize agent behavior. |
| handle_parsing_errors | Boolean | Input parameter. Whether the agent should attempt to recover from input parsing errors. Default: `true`. |
| code_timeout | Integer | Input parameter. Maximum execution time in seconds for each code step. Default: `60`. Range: 10–300. |
| verbose | Boolean | Input parameter. Enable verbose logging. Default: `true`. |
| chat_history | Data | Input parameter. Chat history for multi-turn conversations. |
| agent_description | String | Input parameter. Description of the agent shown when used in Tool Mode. |

## See also

* [File Processing bundle](./bundles-files-ingestion) — file ingestion and retrieval components designed to work with Code Agents
* [OpenDsStar GitHub repository](https://github.com/IBM/OpenDsStar)
* [smolagents documentation](https://huggingface.co/docs/smolagents)