> For the complete documentation index, see [llms.txt](https://docs.ebbot.ai/ebbot-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/workflow-patterns.md).

# Easit workflow patterns

These patterns show how the trigger, actions and Automations nodes fit together. Adapt field names, categories and knowledge sources to the connected Easit environment. The examples are not ready-made customer policies.

### First prove the read path

Start with **Incoming Easit Item → Get Easit Item (GraphQL)** and nothing else. Map the trigger's `item.internal_id` into **Item Id**, save a test ticket in Easit and open the run:

1. The trigger output has a non-null `item.internal_id`.
2. Get Easit Item returns the ticket you saved: check `item.subject`, `item.description` and `item.module_id`.
3. If the output shows a contact or another ticket, the Item Id is mapped from `item.sequence_id`. See [Two ids](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations.md#two-ids-internal-id-and-ticket-number).

Add knowledge retrieval, LLM nodes and writes only after this works. It separates export and connection problems from prompt problems.

### Control the data passed on

Pass only the ticket content each step needs. `item.text` contains every non-empty field, which can include personal data and internal fields; for a knowledge search, `item.subject` and `item.description` are usually enough. Apply your organization's rules for processing, approved services and retention to run traces as well.

Treat ticket text and retrieved articles as untrusted content. They must not change workflow instructions, choose recipients or decide which fields are written. Keep write targets fixed in the action node, and constrain values to those the business process allows.

### Suggest a knowledge article

Goal: when a ticket arrives, find the best matching knowledge article and write it to a field on the ticket.

```
Incoming Easit Item
  → Get Easit Item (GraphQL)          Item Id: trigger item.internal_id
  → LLM node                           RAG on the ticket text, structured output
  → Condition                          structuredOutput.match is true
  → Set Easit Item Field               Item Id: trigger item.internal_id
```

#### 1. Search with the ticket text only

In the LLM node's **RAG** section, set **Search Query** to the ticket's own text, for example `item.subject` and `item.description` from Get Easit Item. Do not add instructions such as "find the matching article and take its title" to the query: the query is used for similarity search, so instructions make results worse.

**Number of documents** controls how many articles reach the LLM. With 5, the LLM gets the five articles most similar to the ticket, and its job is to pick one. Open a run and expand **Sources → View data** to see which fields each article has in your knowledge source, for example a title, the solution text and the article's Easit id.

#### 2. Make the LLM choose and return fields

The prompt must contain the ticket and say what to return. The retrieved articles are added automatically; they need no placeholder.

> You receive a support ticket and up to five knowledge articles. Choose the article whose solution best answers the ticket. If none fits, set match to false and leave the other fields empty.
>
> Ticket: {{ticket text}}

Insert the ticket text with the variable picker (Ctrl/Cmd + +). Then add a schema under **Structured Output**:

```json
{
  "type": "object",
  "properties": {
    "match": {
      "type": "boolean",
      "description": "true if one of the articles answers the ticket, otherwise false"
    },
    "title": {
      "type": "string",
      "description": "Title of the chosen article, exactly as in the source. Empty string if match is false"
    },
    "article_id": {
      "type": "string",
      "description": "Easit id of the chosen article, digits before the colon only (1628373 for 1628373:2). Empty string if match is false"
    },
    "reason": {
      "type": "string",
      "description": "One short sentence on why the article was chosen"
    }
  },
  "required": ["match", "title", "article_id", "reason"],
  "additionalProperties": false
}
```

Each `description` acts as an instruction to the model. Rename the fields and match the descriptions to what your knowledge source contains. With structured output enabled, the node's plain `text` output is no longer available; later nodes use `structuredOutput` and `sources`.

#### 3. Write only when there is a match

Add a condition on the LLM node's `structuredOutput.match`. In **Set Easit Item Field**, map **Item Id** to the trigger's `item.internal_id`, choose the target field from the dropdown, and map **Value** to `structuredOutput.title` for a text field. If the field links to an item, it needs an id rather than a title; confirm with Ebbot how that field type is written before mapping `article_id`.

Check the write's `success` output. To verify, open the ticket in Easit after a test run.

### Classify and set status, category or priority

Fixed-choice fields take data-source ids. Two ways to get a valid id:

* **Fixed list in the prompt.** Run [List Easit Field Values](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/actions/list-field-values.md) once in a test workflow, copy the `label (id)` pairs into the prompt and an `enum` in the structured output schema. Simple to reason about, but update it when the Easit list changes.
* **Look up at runtime.** Give the LLM node the **List Easit Field Values** and **Set Easit Category** tools and instruct it to find the id first, then set it. It stays current with the Easit list, but the LLM performs the write; restrict the tools to that field.

Map the chosen id into [Set Easit Category](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/actions/set-item-category.md), [Set Easit Status](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/actions/set-item-status.md) or [Set Easit Priority](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/actions/set-item-priority.md). A label in place of an id fails the write.

### Reply by email

[Send Easit Email](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/actions/send-easit-email.md) works only after the Easit administrator has created the mail fields and the mail rule, and the field names are entered on the connection. Test with a controlled mailbox first.

* Draft the reply from content suitable for the customer, and check it before sending. Route uncertain cases to a person, for example by writing the draft to a field or with [Record Easit Ticket Message (No Email)](https://docs.ebbot.ai/ebbot-docs/integrations/automations-integrations/integration-setup-guides/easit-integration-in-automations/actions/respond-to-ticket.md), which sends nothing.
* `success` confirms the fields were written and the mail rule was triggered, not that the email arrived.
* Retrying the action can send the email again.
* Customer replies reach the ticket only if Easit imports inbound mail, for example with Email2GO.

### Avoid self-triggering loops

Every write by the workflow saves the ticket in Easit. If the export rule fires on every save, the workflow starts itself again. Ask the Easit administrator to set the rule so that changes made by the workflow do not fire it, for example by firing only on an event that happens once, such as ticket creation, or on a field that only people change.

Then save one test ticket and confirm exactly one run in **Logs**. After the workflow's write, confirm that no second run started.

### Structured output is a node contract

The LLM node's structured output is separate from this integration's action outputs. Its schema is yours to define, and later nodes see only the fields in it. See [LLM Nodes](https://docs.ebbot.ai/ebbot-docs/core-capabilities/automations/workflow-nodes/llm-nodes).
