ASL (Agent Structure Language)

ASL (Agent Structure Language) — an internal DSL(Domain Specific Language) in Python for building agents based on bridgic-core. It provides a declarative, intuitive syntax for defining complex workflow graphs and concurrent automata, making it easier to build and maintain agent-based applications.

Features

• Declarative Syntax: Define workflows using intuitive Python syntax with context managers
• Rich Operators: Use +, ~, >>, and & operators to express complex dependencies
• Fragment Support: Group workers into reusable logical fragments
• Nested Graphs: Support for hierarchical graph structures
• Component Reuse: Reuse existing automata as sub-graphs
• Dynamic Workers: Lambda-based dynamic worker generation for concurrent automata

Installation

pip install bridgic-asl

Quick Start

Basic Example

from bridgic.asl import ASLAutoma, graph

def add_one(x: int):
    return x + 1

def add_two(x: int):
    return x + 2

class MyGraph(ASLAutoma):
    with graph as g:
        a = add_one
        b = add_two
        
        +a >> ~b  # a is the start worker, b depends on a and is the output

# Run the graph
graph = MyGraph()
result = await graph.arun(x=1)  # result: 3 (1+1+2)

Operators Explained

• + marks a worker as a start worker (entry point)
• ~ marks a worker as an output worker (exit point)
• >> defines a dependency (left must complete before right executes)
• & groups multiple workers together

Fragment Example

from bridgic.asl import ASLAutoma, graph

class MyGraph(ASLAutoma):
    with graph as g:
        a = worker1
        b = worker11
        c = worker12
        d = worker5
        e = worker6
        
        # Create fragments
        fragment_1 = +(a & b & c)  # Fragment with start workers
        fragment_2 = (d & e)       # Regular fragment
        
        # Chain fragments together
        fragment_1 >> fragment_2 >> ~merge

Nested Graphs

class MyGraph(ASLAutoma):
    with graph as g:
        a = SubGraph2()
        b = worker2
        
        +a >> b
        
        # Nested graph
        with graph as sub_graph:
            x = worker1
            y = worker2
            +x >> ~y

Configuration and Parameter Injection

from bridgic.asl import ASLAutoma, graph, Settings, Data
from bridgic.core.automa.args import ArgsMappingRule, From

class MyGraph(ASLAutoma):
    with graph as g:
        a = worker1
        b = worker2 *Settings(args_mapping_rule=ArgsMappingRule.AS_IS)
        c = merge *Data(y=From("a"))  # Inject parameter from worker 'a'
        
        +a >> b >> ~c

Concurrent Automa with Dynamic Workers

from bridgic.asl import ASLAutoma, concurrent, ASLField, Settings
from bridgic.core.automa.args import ResultDispatchingRule

class MyConcurrent(ASLAutoma):
    with concurrent(subtasks=ASLField(list, dispatching_rule=ResultDispatchingRule.IN_ORDER)) as sub_concurrent:
        dynamic_logic = lambda subtasks: (
            tasks_done *Settings(key=f"tasks_done_{i}")
            for i, subtask in enumerate(subtasks)
        )

Core Concepts

ASLAutoma

The main class that extends GraphAutoma from bridgic-core. Define your agent by inheriting from ASLAutoma and using the graph or concurrent context managers.

Graph vs Concurrent

• graph: Creates a GraphAutoma for workflow execution
• concurrent: Creates a ConcurrentAutoma for parallel task execution

Settings

Configure worker behavior:

worker *Settings(
    key="custom_key",
    args_mapping_rule=ArgsMappingRule.AS_IS,
    result_dispatching_rule=ResultDispatchingRule.IN_ORDER
)

Data

Inject default parameter values and modify function signatures:

worker *Data(param1=value1, param2=From("other_worker"))

Requirements

• Python >= 3.9
• bridgic-core >= 0.1.5.dev1