AWEL (Agentic Workflow Expression Language)

AWEL is a domain-specific language designed specifically for building LLM application workflows. It lets you compose complex AI pipelines as directed acyclic graphs (DAGs) using a set of built-in operators.

Why AWEL?

Traditional LLM application development involves scattered API calls, fragile glue code, and hard-to-maintain pipelines. AWEL solves this by providing:

Declarative DAGs — Define what your pipeline does, not how to wire it
Reusable operators — Compose from built-in and custom operators
Stream-native — First-class streaming support for real-time responses
Visual editor — AWEL Flow in the Web UI for no-code workflow building

How it works

mermaid

flowchart LR
    T["Trigger (HTTP/Schedule)"] --> A["Operator A"]
    A --> B["Operator B"]
    A --> C["Operator C"]
    B --> D["Join Operator"]
    C --> D
    D --> E["Output"]

An AWEL pipeline consists of:

Trigger — Entry point (HTTP request, schedule, or manual invocation)
Operators — Processing nodes that transform data
DAG — The graph connecting triggers and operators

Core operators

Operator	Description	Use case
MapOperator	Transform each input item	Data formatting, API calls
ReduceOperator	Aggregate multiple inputs into one	Summarization, collection
JoinOperator	Merge results from parallel branches	Multi-source aggregation
BranchOperator	Route input to different paths	Conditional logic
StreamifyOperator	Convert batch to stream	Real-time processing
UnstreamifyOperator	Convert stream to batch	Collecting stream results
TransformStreamOperator	Transform items in a stream	Stream filtering/mapping
InputOperator	Provide initial input to a DAG	Pipeline entry data

Quick example

A minimal AWEL workflow that takes a user question and generates an LLM response:

python

from dbgpt.core.awel import DAG, MapOperator, InputOperator

with DAG("simple_chat") as dag:
    input_node = InputOperator(input_source="user_question")
    llm_node = MapOperator(map_function=call_llm)
    input_node >> llm_node

AWEL Flow (visual editor)

The Web UI includes a drag-and-drop AWEL Flow editor where you can:

Build workflows visually by connecting operator nodes
Configure each operator's parameters in a sidebar
Test and debug flows in real-time
Save and share flow templates

Access it from the Web UI sidebar under AWEL Flow.

What's next

AWEL Tutorial — Step-by-step learning path
AWEL Cookbook — Practical recipes for common patterns
AWEL Flow Usage — Using the visual editor