> ## Documentation Index
> Fetch the complete documentation index at: https://docs.adopt.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Pipelines
> Scheduled jobs that pull data from your systems, transform it, and land it in the Data Store for agents to use.
Agents work best when the data they need is already clean, structured, and waiting for them — not fetched live in the middle of a conversation. Pipelines are what make that true. A Pipeline connects to your external systems on a schedule, extracts and transforms the data, and lands it in the [Data Store](/accounting/data-store), so agents read pre-computed results instead of hammering raw APIs mid-run.
## What Are Pipelines?
A Pipeline is pure data infrastructure. On a schedule or trigger, it reaches out through [Connectors](/accounting/connectors), pulls data, reshapes it, and writes it to the Data Store. It doesn't make decisions — it prepares inputs.
Pipelines are the extract-and-transform engine of the platform — integration, extraction, and transformation that hands clean, structured data to everything downstream.
A crucial detail: **a Pipeline is always defined within a [Skill](/accounting/skills).** The pipeline is that skill's mechanism for bringing in fresh data. Because skills compose into agents, a pipeline is associated with an agent only *through* its skill — never directly. Pipelines get their own management surface under the **Adopt Primitives** group, but they remain bound to a skill.
## How to Access Pipelines
Click **Pipelines** in the left navigation, under the **Adopt Primitives** group. This opens the pipeline library.
## Inside the Pipelines List
The library lists every pipeline as a card. Tabs let you switch between **All Pipelines** and an **Overview**, filter chips separate **Active** from **Paused**, and dropdowns filter by schedule or workstream.
Each card surfaces everything you need to understand a pipeline at a glance:
| Field | What it shows |
| ------------------------ | ------------------------------------------------------------------------------ |
| **Name & status** | The pipeline's name and whether it's **Active** or **Paused**. |
| **Description** | What the pipeline does. |
| **Type** | Whether it's an *Agent pipeline* or driven by a *Connector*. |
| **Source → destination** | The source system and the destination — almost always the Internal Data Store. |
| **Schedule** | How it runs: *Manual*, or on a cadence like *Every 15 minutes*. |
| **Workstream** | The client this pipeline's data belongs to. |
| **Updated by** | Who last changed it, and when. |
## Building a Pipeline
Click **New Pipeline** to open the builder. It walks you through four steps.
| Step | What you configure |
| --------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Source** | Name the pipeline, associate the **Skills** and **Agent** it belongs to, and choose the source — one or more Connectors, or the Internal Data Store itself. |
| **Workstreams** | Bind the run's data to a client, so it lands in the right workstream. |
| **Schedule** | Set how it runs: on a cadence, on an event, or on demand. |
| **Describe** | Describe the extract-and-transform logic that cleans, joins, or enriches the data in flight. |
Whatever the source, the destination is the same: **data syncs to the Internal Data Store**, where the structured result becomes available for dashboards, agents, and export.
## How Pipelines Behave
Pipelines run **independently of agent execution**. They pre-compute data on a schedule rather than blocking a conversation — so when an agent runs, the data it needs is already sitting in the Data Store. A skill can own one or more pipelines; their output lands in the Data Store, and the skill's agents read it from there. A pipeline never feeds an agent directly.
## Why Pipelines Matter
Pipelines are what keep agents fast and reliable. Instead of an agent making a dozen slow, fragile API calls every time someone asks for a flux report, a pipeline has already pulled and cleaned the ledger overnight. The agent just reads the result. That separation — heavy data work on a schedule, light reasoning at runtime — is what lets the platform handle real accounting volume without falling over.
## Next Steps
1. [See where pipeline data lands in the Data Store](/accounting/data-store)
2. [Understand the skill that owns a pipeline](/accounting/skills)
3. [Review Pipeline behaviour in the wider data flow](/accounting/how-it-connects)