Patterns you can recreate in the workflow editor—inputs, webhooks, schedules, and common step chains—with sample JSON for HTTP triggers.

Example workflows

These are patterns, not one-click templates: you add steps on the canvas, connect them left-to-right, and fill each step’s panel. For starter graphs and real-world scenarios (support, sales, schedules, webhooks), see Templates and real use cases. Use Workflows overview for step types, Triggering workflows for webhook POST details, and Wisebot CLI and HTTP API for the full wisebot tool (not only workflows).

How examples map to the editor

Input steps define field names (for example topic, emailBody). Those names become JSON keys in a webhook body or Run now form.
Output picks which upstream step’s result is the primary workflow result (what you see in run history).
Webhook runs use POST …/workflows/trigger?workflowId=… with your secret; see the trigger section in the editor.

1. Summarize text (manual or webhook)

Goal: Take a short message and produce a concise summary.

Order	Step	Notes
1	Input	One text field, e.g. `rawText`.
2	LLM	Prompt: “Summarize the following in three bullets…” and reference the input (per your UI—usually templating into the prompt).
3	Output	Point at the LLM step.

Sample webhook body

{ "rawText": "Long customer email pasted here…" }

2. Classify and branch with Transform + Log

Goal: Call an external classifier or LLM, then pull one field for monitoring.

Order	Step	Notes
1	Input	e.g. `ticket` (text).
2	LLM	Ask for strict JSON output—for example fields `priority` (low, med, or high) and `reason` (short text).
3	Transform	Dot path into the parsed result if needed (e.g. `priority` if your runtime exposes it flat).
4	Log	Optional: log priority or reason for ops.
5	Output	Final structured result or the LLM step.

Use Transform whenever you need a single field from nested JSON returned by an upstream step.

3. HTTP callback after enrichment (notify another system)

Goal: Enrich with an LLM, then POST a payload to your CRM or internal API.

Order	Step	Notes
1	Input	Fields your webhook sends (`leadId`, `notes`, …).
2	LLM	Draft a short “summary for CRM” from `notes`.
3	HTTPS	POST to your API path with Input placeholders in the URL (for example `/leads/<leadId>/note` using the placeholder syntax shown in the editor).
4	Output	HTTPS response or LLM summary—your choice.

Ensure the target URL is HTTPS and allowed by your organization’s networking rules.

4. Knowledge-assisted answer (KB preflight → LLM)

Goal: Ground an answer in your knowledge base before the model writes the final text.

Order	Step	Notes
1	Input	e.g. `question` (text).
2	KB preflight	Attach the agent whose knowledge files you want; the step pulls relevant snippets.
3	LLM	Prompt: answer using only the provided context; cite gaps if missing.
4	Output	LLM step.

This matches how agents use knowledge, but as a single scripted run with a stored result.

5. Scheduled digest (schedule + inputs with defaults)

Goal: Run on a timer (e.g. daily) to produce a short report.

Order	Step	Notes
1	Input	Optional fields; for a pure schedule you can use fixed prompts in later steps.
2	KB preflight or LLM	e.g. “List top 3 open themes from yesterday’s…” if you pass context via inputs or static instructions.
3	Output	Final text.

Enable Scheduled runs in the workflow settings and pick an interval. Scheduled runs may use empty or default inputs depending on how you configured Input steps.

6. Agent tool on an existing conversation (support automation)

Goal: Run one tool the same way as in chat—e.g. search or escalate—for a real thread.

Order	Step	Notes
1	Input	Must supply identifiers your deployment expects: typically thread / conversation id and session / contact context, plus tool arguments. Exact field names match what Agent tool expects in the editor.
2	Agent tool	Choose the tool (e.g. `searchTool`, `escalateConversation`).
3	Output	Tool result.

Important: These steps do not create conversations—they operate on threads that already exist and belong to your organization. Use this when an external system (or a prior chat step) already created the thread.

7. MCP: create a task in Notion or Slack

Goal: Use a connected MCP integration to perform an action in another product.

Order	Step	Notes
1	Input	e.g. `title`, `description` from the webhook.
2	MCP tool	Pick the integration and tool name; pass JSON arguments (map from inputs or static JSON).
3	Output	MCP response.

Connect the integration under Integrations first. If the tool fails, the run shows the same error details as elsewhere in the product.

8. Sandbox code (Python / Node / bash)

Goal: Run a small script with inputs and upstream context (when your workspace has Code enabled).

Order	Step	Notes
1	Input	Data the script needs.
2	Code	Choose runtime; script receives inputs, trigger metadata, and payload as documented in the editor.
3	Output	Code stdout/result mapping (per your step configuration).

If Code is unavailable, your administrator may need to enable sandbox execution for the deployment.

Background: E2B sandboxes for AI agents explains why ephemeral, isolated runtimes matter when models drive Code and tool steps—and how that relates to Workflows overview.

Templates and real use cases — Starter graphs and business scenarios
Workflows overview — Triggers, limits, and full step list
Triggering workflows (HTTPS & CLI) — curl, wisebot workflow trigger, Node/Python POST examples
Wisebot CLI and HTTP API — all wisebot commands and dashboard /api/v1/… usage
Built-in tools — Names for Agent tool
MCP overview — Connecting MCP tool integrations
E2B sandboxes for AI agents — isolation patterns for Code and tools

Example workflows

On this page