AI Tasks

Leverage AI for summarization, investigation, classification, and routing.

llm.summary

Display Name: AI Summary

Summarize text, logs, or data using AI. Pass any content and receive a concise summary highlighting key points.

Parameters

Name	Type	Required	Description
message	string	Yes	The text, logs, or data to summarize. Supports template expressions to reference previous task outputs.

Output

Name	Type	Description
data	string	AI-generated summary.
conversation_id	string	NuBi conversation ID (for follow-up).
session_id	string	NuBi session ID.

---

llm.investigate

Display Name: AI Investigation

Ask AI to analyze and investigate a problem. The AI will research the issue using available tools and context, returning detailed findings and recommendations.

Parameters

Name	Type	Required	Description
message	string	Yes	Description of the problem or question to investigate. Supports template expressions.

Output

Name	Type	Description
data	string	Investigation findings and recommendations.
conversation_id	string	NuBi conversation ID.
session_id	string	NuBi session ID.

---

llm.event_investigate

Display Name: AI Event Investigation

Ask AI to investigate a specific event or alert. Designed for event-triggered workflows — automatically analyzes the event context, checks related resources, and provides root cause analysis.

Parameters

Name	Type	Required	Description
message	string	Yes	Event or alert details. Typically pass {{ event }} for webhook-triggered workflows.

Output

Name	Type	Description
data	string	Root cause analysis and recommendations.
conversation_id	string	NuBi conversation ID.
session_id	string	NuBi session ID.

---

llm.nubi

Display Name: Ask NuBi

Ask NuBi (Nudgebee's AI assistant) to investigate an issue or answer a question. NuBi has access to your infrastructure context including K8s clusters, services, and events.

Parameters

Name	Type	Required	Description
message	string	Yes	Question or issue to ask NuBi about.

Output

Name	Type	Description
data	string	NuBi's response.
conversation_id	string	NuBi conversation ID.
session_id	string	NuBi session ID.

---

llm.classify

Display Name: AI Classifier

Use AI to categorize input into one of several predefined options. Useful for routing decisions based on content analysis.

Parameters

Name	Type	Required	Description
prompt	string	Yes	The user query or context to evaluate.
options	array	Yes	List of options, each with name (branch identifier) and description (what it represents).

Output

Name	Type	Description
selected_branch	string	The name of the selected option.

---

llm.router

Display Name: AI Router

Use AI to classify input and automatically route to the correct branch of tasks. Define multiple branches with descriptions — the AI selects which branch to execute.

Parameters

Name	Type	Required	Description
prompt	string	Yes	The input to classify and route.
branches	array	Yes	List of branches, each with name, description, and tasks (list of task definitions to execute).

Output

Name	Type	Description
selected_branch	string	Name of the branch that was selected and executed.

---

llm.a2a_call

Display Name: AI Agent Call

Call an external AI agent via JSON-RPC 2.0. Use this to integrate with third-party AI agents or services that expose an agent-to-agent (A2A) compatible endpoint.

Parameters

Name	Type	Required	Description
url	string	Yes	External agent endpoint URL.
method	string	Yes	JSON-RPC method name (e.g., agent.chat).
params	any	No	Parameters for the JSON-RPC call (JSON object).
headers	object	No	Custom headers (e.g., Authorization).

Output

Name	Type	Description
result	any	Result of the JSON-RPC call.
id	string	Request ID echo.
jsonrpc	string	JSON-RPC version.

---

llm.mcp_call

Display Name: MCP Call

Call a tool on an external MCP (Model Context Protocol) compatible server.

Parameters

Name	Type	Required	Default	Visibility	Description
connection_mode	string	Yes	integration	Always	How to connect to the MCP server. Options: integration (use a saved MCP integration) or direct (provide URL and auth inline).
integration_id	integration	Conditional	—	connection_mode = integration	Select an MCP integration to use. Manage integrations under Settings > Integrations.
url	string	Conditional	—	connection_mode = direct	The URL of the MCP server. Tip: Save connection details as an MCP integration under Settings > Integrations for reuse.
tool_name	string	Yes	—	Always	The name of the tool to invoke. The tool name dropdown is dynamically populated — it connects to the MCP server (via integration or direct URL) and fetches available tools using the tools/list MCP method.
arguments	object	No	{}	Always	Tool-specific input arguments.
headers	object	No	—	connection_mode = direct	Custom HTTP headers to include in the request (e.g., Authorization).
auth_type	string	No	"" (none)	connection_mode = direct	Authentication type. Options: "" (none) or oauth2. For bearer, basic, or API key authentication, use the headers parameter directly instead.
oauth_token_url	string	Conditional	—	auth_type = oauth2 AND connection_mode = direct	OAuth 2.0 token endpoint URL.
oauth_client_id	string	Conditional	—	auth_type = oauth2 AND connection_mode = direct	OAuth 2.0 client ID.
oauth_client_secret	string	Conditional	—	auth_type = oauth2 AND connection_mode = direct	OAuth 2.0 client secret. Encrypted at rest.
oauth_scope	string	No	—	auth_type = oauth2 AND connection_mode = direct	OAuth 2.0 scopes (space-separated).
oauth_audience	string	No	—	auth_type = oauth2 AND connection_mode = direct	OAuth 2.0 audience identifier. Required by some providers like Auth0.
timeout	string	No	60s	Always	Request timeout (e.g., 30s, 2m).

Connection Modes

Integration Mode (Recommended)

• Uses a pre-configured MCP integration from Settings > Integrations.
• Connection details, authentication, and credentials are managed centrally.
• Supports both direct HTTP connections and vm_agent connections (routed through forager agent for on-prem MCP servers).
• All auth types supported: none, bearer, basic, API key/custom header, OAuth 2.0.

Direct Mode

• Provide the MCP server URL and authentication inline in the task configuration.
• Useful for quick testing or one-off connections.
• Supports custom headers and OAuth 2.0 authentication.

How It Works

The task uses the MCP Streamable HTTP transport with JSON-RPC 2.0:

1. Sends an initialize request to establish a session (protocol version 2024-11-05).
2. Sends an initialized notification.
3. Sends a tools/call request with the selected tool name and arguments.
4. Returns the tool's response (content array and isError flag).

The task supports both immediate JSON responses and Server-Sent Events (SSE) streams from the MCP server.

Output

Name	Type	Description
content	array	Content returned by the tool.
is_error	boolean	Whether the tool execution resulted in an error.