mirror of
https://github.com/langflow-ai/langflow.git
synced 2026-07-25 00:59:07 +08:00
71fbf5ff00
* version-1.9.0-docs * [autofix.ci] apply automated fixes * [autofix.ci] apply automated fixes (attempt 2/3) --------- Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
96 lines
6.6 KiB
Plaintext
96 lines
6.6 KiB
Plaintext
---
|
|
title: CUGA
|
|
slug: /bundles-cuga
|
|
---
|
|
|
|
import Icon from "@site/src/components/icon";
|
|
import PartialParams from '@site/docs/_partial-hidden-params.mdx';
|
|
|
|
<Icon name="Blocks" aria-hidden="true" /> [**Bundles**](/components-bundle-components) contain custom components that support specific third-party integrations with Langflow.
|
|
|
|
:::warning Model Provider Limitations
|
|
The **CUGA** component only supports **OpenAI** and **watsonx** models. To use other model providers, use the core [**Agent** component](/agents) instead.
|
|
:::
|
|
|
|
The **CUGA (ConfigUrable Generalist Agent)** component is an advanced AI agent that executes complex tasks using tools, optional browser automation, and structured output generation.
|
|
|
|
The **CUGA** component can be used in flows in place of an [**Agent** component](/agents).
|
|
Like the core **Agent** component, the **CUGA** component can use tools connected to its **Tools** port, and can be used as a tool itself.
|
|
It also includes some additional features:
|
|
|
|
* Browser automation for web scraping with [Playwright](https://playwright.dev/docs/intro).
|
|
To enable web scraping, set the component's `browser_enabled` parameter to `true`.
|
|
* Load custom instructions for the agent to execute.
|
|
To use this feature, use the component's **Instructions** input to attach markdown files containing agent instructions.
|
|
|
|
For more information, see the [CUGA project repository](https://github.com/cuga-project/cuga-agent).
|
|
|
|
## Use the CUGA component in a flow
|
|
|
|
For demonstration purposes, the following example modifies a template flow to use the **CUGA** component.
|
|
|
|
|
|
|
|
1. Create a flow based on the **Simple Agent** template, and then replace the **Agent** component with the **CUGA** component.
|
|
2. Connect an [**MCP Tools** component](/mcp-client) and a [**Calculator** component](/calculator) to the **CUGA** component's **Tools** port, and then connect the **MCP Tools** component to any MCP server.
|
|
This example connects to a server containing sales data for a business organization.
|
|
3. Add a [**Read File** component](/read-file), and then connect it to the **CUGA** component's **Instructions** port.
|
|
Alternatively, click <Icon name="Scan" aria-hidden="true" /> **Edit text** to open the **Edit text content** pane, and enter your instructions directly into the **CUGA** component.
|
|
4. Create a Markdown file on your computer called `instructions.md`, and then insert the following content.
|
|
It's important to clearly format the document with `## Plan` and `## Answer` for the agent to understand your instructions.
|
|
```markdown
|
|
## Plan
|
|
|
|
- Break down complex queries into subtasks
|
|
- Prioritize information gathering before execution
|
|
- Consider dependencies between actions
|
|
- Validate intermediate results before proceeding
|
|
|
|
## Answer
|
|
|
|
- Provide concise summaries with key findings
|
|
- Include relevant data points and metrics
|
|
- Cite sources when using MCP tool results
|
|
- Use clear structure and formatting for readability
|
|
```
|
|
5. In the **Read File** component, click <Icon name="Plus" aria-hidden="true"/> **Add File**, and then attach your `instructions.md` file.
|
|
|
|
6. Open the **Playground**, and then ask the agent a question that could use your connected MCP server.
|
|
This example asked about the sales data provided by the MCP Server, such as `Which accounts are available?`.
|
|
The agent describes the tool calls it makes and then returns an answer according to the instructions.
|
|
For example, the list of available accounts is very large, but the **CUGA** component returns a concise summary as requested in the policy.
|
|
```
|
|
Based on the available data, here are the accounts:
|
|
|
|
Summit Solutions (NY) - Revenue: $1,200,000
|
|
Pacific Ventures (CA) - Revenue: $9,500,000
|
|
Lone Star Corp (TX) - Revenue: $4,500,000
|
|
Mountain Peak Systems (CO) - Revenue: $2,100,000
|
|
Digital Dynamics (CA) - Revenue: $5,500,000
|
|
Cascade Computing (WA) - Revenue: $4,300,000
|
|
Data Flow Systems (CA) - Revenue: $8,900,000
|
|
Rocky Mountain Enterprises (CO) - Revenue: $3,200,000
|
|
Blue Sky Partners (TX) - Revenue: $500,000
|
|
Liberty Manufacturing (PA) - Revenue: $3,400,000
|
|
|
|
This is a partial list; there are more accounts available. The total revenue across all accounts is $210,200,000.
|
|
```
|
|
|
|
## CUGA parameters
|
|
|
|
<PartialParams />
|
|
|
|
| Name | Type | Description |
|
|
|------|------|-------------|
|
|
| agent_llm | Dropdown | Model provider for the agent.|
|
|
| instructions | Multiline String | Custom instructions that define the agent's planning and answers. Can be provided directly or through Markdown files. Formatting is important in order for the agent to understand the instructions. See [Use the CUGA component in a flow](#use-the-cuga-component-in-a-flow). |
|
|
| n_messages | Integer | Number of chat history messages to retrieve. Useful for maintaining context in ongoing conversations identified by `session_id`. Default: `100`. |
|
|
| format_instructions | Multiline String | Template for structured output. |
|
|
| output_schema | Table | When `output_schema` is provided, structured responses are validated against a dynamically built schema. Invalid items are returned alongside validation errors. Fields: `name`, `description`, `type` (`str`, `int`, `float`, `bool`, `dict`), `multiple` (as list).|
|
|
| add_current_date_tool | Boolean | If true, adds a tool that returns the current date. Default: `true`. |
|
|
| lite_mode | Boolean | Set to `true` to enable CugaLite mode for faster execution when using a smaller number of tools. Default: `true`. |
|
|
| lite_mode_tool_threshold | Integer | The threshold to automatically enable CugaLite. If the CUGA component has fewer tools connected than this threshold, CugaLite is activated. Default: `25`. |
|
|
| shortlisting_tool_threshold | Integer | The threshold for tool shortlisting. When the total number of tools exceeds this threshold, the CUGA component enables its `find_tools` feature to filter tools down to a smaller subset before making tool selection decisions. This helps reduce token usage and improve performance when working with large numbers of tools. Default: `35`. |
|
|
| decomposition_strategy | Dropdown | Strategy for task decomposition. `flexible` allows multiple subtasks per app. `exact` enforces one subtask per app. Default: `flexible`. |
|
|
| browser_enabled | Boolean | Enable a built-in browser for web scraping and search. Allows the agent to use general web search in its responses. Disable (`false`) to restrict the agent to the context provided in the flow. Default: `false`. |
|