How to Add AI Summaries to monday.com Items

Connect monday.com to AI: summarise each item's updates into a short status line, suggest the next step and flag blockers, and write the result back to a column on the board every morning.

What This Integration Does

Busy monday.com boards hide their real status in long update threads. Nobody reads forty updates before a stand-up. This Spojit workflow reads the latest updates on every active item, asks an AI model to summarise them in one or two sentences, pick the next step and say whether the item is blocked, and writes those answers into columns on the board. The board then tells you what is going on at a glance, and you can sort or filter by the "Blocked" column.

The AI runs inside Spojit with the models Spojit provides, so you do not need an OpenAI, Llama or other AI account (if you do want a specific provider, see the Tips). The workflow runs once a day, touches only items in the groups you choose, and overwrites the summary columns each time, so re-running is safe and the summary always reflects the latest updates.

Prerequisites

A Monday.com connection in Spojit (personal API token).
On the board: a Long text column for the summary, a Text column for the next step, and a Status column with labels Blocked and On track. Get their column IDs with the get-board tool.
AI credits on your Spojit plan for one short call per item.

Step 1: Schedule Trigger

Add a Trigger node with type Schedule and cron 0 7 * * 1-5 in your team's timezone, so summaries are fresh before the morning meeting.

Step 2: Read the Items and Their Updates

Add a Monday.com step with raw-graphql, which fetches items and their recent updates in one call:

query {
  boards(ids: [YOUR_BOARD_ID]) {
    items_page(limit: 50) {
      items {
        id
        name
        group { title }
        updates(limit: 15) { text_body created_at creator { name } }
      }
    }
  }
}

Name the output board. For boards with more than 50 active items, page with the cursor that items_page returns. Add a Loop in ForEach mode over the items, with a Condition that skips items in groups you do not want summarised (such as "Done").

Step 3: Summarise with AI

Inside the loop, add a Connector node in Agent Mode on the Text Tools connector. An empty tool selection does not mean "no tools" (it gives the agent every tool on the connector), so use Text Tools, whose tools only reshape text and cannot change any system. Prompt:

You summarise project items for a team stand-up.
Item: {{ item.name }}
Recent updates, newest first:
{{ item.updates }}

Write a summary of the current status in at most two sentences, the single next step,
and whether the item is blocked (waiting on someone or something outside the team).
Use only what the updates say.

Add a Response Schema with summary (string), next_step (string) and blocked (boolean). Name the output ai. A small, fast model is enough for this task.

Step 4: Write the Results Back

Add a Monday.com step with update-item: boardId your board, itemId {{ item.id }}, and columnValues:

{
  "long_text_summary": { "text": "{{ ai.summary }}" },
  "text_next_step": "{{ ai.next_step }}",
  "status_health": { "label": "(Blocked when ai.blocked is true, otherwise On track)" }
}

Use a Condition on {{ ai.blocked }} with one update-item per branch to set the right label. Replace the keys with your column IDs.

Step 5: Post Blockers to the Team (Optional)

Collect the blocked items and, after the loop, send one Slack message listing them with their next steps, or post an update on each blocked item with create-update so its owner gets a monday.com notification.

Tips

Using OpenAI or another provider: if you must use a specific provider, replace Step 3 with an HTTP Requests call to that provider's API (see the "How to Call the OpenAI API from a Workflow" tutorial in this help center). The rest of the workflow is unchanged.
Weekly digest: run a second workflow on Fridays that summarises the whole board into one paragraph for leadership, using the item summaries as input.
Use files too: if items carry documents, add a Knowledge step or PDF Tools extract-text so the summary can mention what the documents say.
Columns, not updates: writing summaries to columns keeps the update thread clean and lets you sort and filter by the AI's assessment.

Common Pitfalls

Long text column format: long text columns expect { "text": "..." }, while plain text columns take a string. The wrong shape makes update-item fail.
Status labels must exist: create Blocked and On track on the status column first, or the update fails.
Too many updates: sending every update in a long thread costs more and adds noise. Fifteen recent updates is usually plenty.
Private boards: the API token only sees boards its user can see. Add that user to private boards you want summarised.

Testing

Point the query at a test group with three items: one with recent progress, one waiting on a supplier, and one with no updates. Run the workflow manually and check the columns: the second should be Blocked, and the third should say there is no recent activity rather than inventing one. Then turn on the schedule.

Learn More

Ready to build this?

Start on the free plan: 500 AI credits a month, no card needed.

Start free