> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rippit.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Pipelines

> A pipeline is a chain of steps over your data that keeps running on its own: it enriches every record, looks things up, researches the web, rolls results up, and acts on what it finds.

The analysis worth doing is usually worth repeating. You want every new conversation labeled, not only last month's. You want the churn report every Monday, not the one week someone remembered to ask. A pipeline keeps that work going: you describe the outcome once, Rippit builds the chain of steps that produces it, and from then on it runs on its own, on every new record, on a schedule, or both.

You don't assemble anything by hand. Describe what you want in Rippit chat, or to your own AI assistant through the [Rippit MCP server](https://docs.rippit.com/mcp-server/overview.md), and the pipeline is built in that conversation, where you can also test it, change it, and put it live.

## Why pipelines

* **One place for the whole job.** Enriching records, bringing in account data, researching the web, summarizing, and sending the result are steps of one pipeline, not separate tools you connect yourself.
* **Shaped to your problem.** You combine the steps the job needs instead of fitting the job to a fixed report.
* **Built from a sentence.** Rippit works out the steps, asks about what it can't infer, and shows you the pipeline before anything runs.
* **Nothing hidden.** Every step writes to a table you can open, query, and chart, so you can see why a pipeline said what it said.
* **Safe to leave running.** You test on a sample, get a cost estimate before going live, and can pause at any time.

Use chat when you want an answer once. Use a pipeline when the same work should keep happening.

## What you can build

A pipeline is made of a few kinds of steps, and you combine them however the job requires. Three shapes cover most of what teams build. A pipeline can be any one of them, or all three at once.

> [card] Enrich every record
>
> Add AI columns to a table: a contact reason on every conversation, a churn-risk score on every call. Rippit fills them for the rows you have and for each new row that arrives.

> [card] Report and alert on a schedule
>
> Every Monday, summarize what changed and post it to Slack. Every hour, check for a spike and alert only when there is one.

> [card] Research the outside world
>
> Search the web, find people, or ask AI models such as ChatGPT a question, for each account or on a schedule, and save every finding as a row.

The steps compose, so one pipeline can carry a job from raw data to a finished action:

```text wrap theme={null}
Label every support conversation with its product area and whether it reports a bug. Look up each customer's plan and account owner in our accounts table. Every weekday at 9am, post yesterday's bugs to #triage, one message per product area, with the enterprise accounts affected and links to the conversations.
```

That one request becomes AI columns on your conversations, a lookup into your accounts table, a daily rollup with one row per product area, an agent run that writes each area's summary, and an action that posts each one to Slack. See [more examples](https://docs.rippit.com/pipelines/examples.md).

## The building blocks

| Building block | What it does |
| - | - |
| [Data source](https://docs.rippit.com/pipelines/steps.md#data-sources) | The table a pipeline reads: conversations from an integration, business data from Salesforce, Snowflake, or the Bulk API, or the output of another pipeline |
| [AI column](https://docs.rippit.com/pipelines/steps.md#ai-columns) | A model reads each row, and the full conversation behind it, and fills in an answer: a label, a score, a yes or no, a summary |
| [Formula](https://docs.rippit.com/pipelines/steps.md#formulas) | Calculates a column from other columns, with no AI cost |
| [Lookup](https://docs.rippit.com/pipelines/steps.md#lookups) | Pulls in matching rows from another table, such as the account behind a ticket |
| [Rollup](https://docs.rippit.com/pipelines/steps.md#rollups) | On a schedule, folds a window of rows into one row per account, agent, topic, or period, with counts and other totals |
| [Agent run](https://docs.rippit.com/pipelines/steps.md#agent-runs) | Runs an agent on each row to investigate, write, or decide, with the tools you give it |
| [External research](https://docs.rippit.com/pipelines/steps.md#external-research) | What an agent run can do beyond your data: search the web, find people, and ask AI models such as ChatGPT a question |
| [Action](https://docs.rippit.com/pipelines/steps.md#actions) | Sends the result somewhere: a Slack post, an email, a new ticket, a CRM update, or any tool on a connected MCP server |

Every step writes its results to a table. That is what makes pipelines flexible: any step can read what an earlier step produced, and you can query, chart, and ask questions about every table along the way.

## How a pipeline runs

Steps run at one of two rhythms, and most pipelines use both.

* **As data arrives.** AI columns, formulas, and lookups run on each new row as its source syncs, and stay current when the row changes.
* **On a schedule.** A rollup runs hourly, daily, weekly, or monthly. The agent runs and actions after it then do their work once for each row the rollup produced, so one row per account means one message per account.

## From idea to live

1. **Describe it.** Say what should happen, on which data, how often, and where the result goes. Rippit asks about anything the data can't answer.
2. **Review the draft.** The pipeline appears as a graph you can inspect. Nothing runs yet.
3. **Test it.** Run every step once on a small sample and read the results.
4. **Go live.** Once you confirm, the pipeline starts and keeps running until you pause it.

[Build, test, and run a pipeline](https://docs.rippit.com/pipelines/build-and-run.md) covers each stage.

## Where pipelines live

Open **Pipelines** in the left sidebar to see every pipeline in your workspace, with its status, its number of steps, and when it last ran. Open one to see its graph and the rows each step produced; [Keep an eye on it](https://docs.rippit.com/pipelines/build-and-run.md#keep-an-eye-on-it) describes the page. The tables a pipeline produces are also listed under **Tables**.

> [card] [Pipeline building blocks](https://docs.rippit.com/pipelines/steps.md)
>
> What each kind of step can do.

> [card] [Build, test, and run](https://docs.rippit.com/pipelines/build-and-run.md)
>
> Describe a pipeline, test it safely, and put it live.

> [card] [Rules and limits](https://docs.rippit.com/pipelines/reference.md)
>
> How steps connect, how to design a pipeline well, and every limit.

> [card] [External research](https://docs.rippit.com/pipelines/research.md)
>
> Web search, people search, and agent search, with the providers you can choose.

> [card] [Examples](https://docs.rippit.com/pipelines/examples.md)
>
> Prompts to start from, for support, revenue, and product teams.

> [card] [Build from your AI assistant](https://docs.rippit.com/pipelines/build-and-run.md#build-from-your-ai-assistant)
>
> Create and manage pipelines from any MCP client.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.