What you need to know

Since BLAST automatically runs tasks concurrently, the only thing you may need to set is constraints:

from blastai import Engine, Constraints

engine = await Engine.create(
    constraints=Constraints(
        allow_parallelism=True,     # Enable/disable parallel execution
        max_concurrent_browsers=4    # Maximum number of parallel browsers
    )
)

​
Task Lifecycle

Every task in BLAST goes through several stages:

  1. Creation: Task is scheduled and assigned a unique ID
  2. Resource Allocation: Browser and other resources are assigned
  3. Execution: Task runs and streams progress updates
  4. Completion: Results are stored and resources are freed

​
Task Relationships

BLAST supports two types of task relationships:

​
Prerequisites

Tasks can depend on other tasks completing first:

from blastai import Engine

async with Engine() as engine:
    # Task B won't start until Task A completes
    task_a = await engine.run("Go to python.org")
    task_b = await engine.run(
        "Click on Documentation",
        previous_response_id=task_a.id  # Set prerequisite
    )

​
Parent/Child Tasks

Tasks can spawn subtasks that run concurrently:

# Parent task can create multiple subtasks
results = await engine.run("Launch subtasks to visit each of these websites: bloomberg.com, wsj.com, and nytimes.com")

​
Task Priorities

BLAST prioritizes tasks in this order:

  1. Tasks with cached results
  2. Tasks with cached execution plans
  3. Subtasks of running tasks
  4. Tasks with paused executors
  5. Remaining tasks (FIFO order)

​
Concurrent Execution

BLAST can execute multiple tasks concurrently when resources allow:

async with Engine(
    constraints=Constraints(
        max_concurrent_browsers=4,  # Allow 4 parallel browsers
        allow_parallelism=True  # Enable parallel execution
    )
) as engine:
    # These tasks will run in parallel
    task1 = engine.run("Search Python docs", stream=True)
    task2 = engine.run("Search JavaScript docs", stream=True)
    task3 = engine.run("Search Rust docs", stream=True)
    
    # Process streams concurrently
    async def process_stream(stream):
        async for update in stream:
            if isinstance(update, AgentReasoning):
                print(update.content)
    
    await asyncio.gather(
        process_stream(task1),
        process_stream(task2),
        process_stream(task3)
    )

​
Resource Management

BLAST automatically manages resources for concurrent tasks:

  • Limits concurrent browser instances
  • Reuses browsers when possible
  • Manages memory usage
  • Handles cleanup on task completion

Monitor resource usage:

metrics = await engine.get_metrics()
print(f"Running tasks: {metrics['tasks']['running']}")
print(f"Active browsers: {metrics['concurrent_browsers']}")
print(f"Memory usage: {metrics['memory_usage_gb']} GB")

​
Task State Management

Track task states through the API:

metrics = await engine.get_metrics()
task_states = metrics['tasks']

print(f"Scheduled: {task_states['scheduled']}")  # Waiting to start
print(f"Running: {task_states['running']}")      # Currently executing
print(f"Completed: {task_states['completed']}")  # Finished tasks

​
Next Steps