Task Context Control

Control how much prior task output each step receives — pass only what the next task needs instead of the full conversation history.

from praisonaiagents import Agent, Task, AgentTeam

processor = Agent(name="Processor", instructions="Process data efficiently")

extract = Task(
    name="extract",
    description="Extract data from source files",
    expected_output="Extracted data summary",
    agent=processor,
    retain_full_context=False,  # Only this task's output flows forward
)

analyse = Task(
    name="analyse",
    description="Analyse trends across all previous outputs",
    expected_output="Trend analysis",
    agent=processor,
    retain_full_context=True,  # Include all prior task outputs
)

team = AgentTeam(agents=[processor], tasks=[extract, analyse], process="sequential")
team.start()

The user runs a sequential team workflow; each step receives either just the prior task output or the full history, depending on retain_full_context.

How It Works

Quick Start

Create tasks with context control

from praisonaiagents import Agent, Task, AgentTeam

processor = Agent(
    name="DataProcessor",
    instructions="Process and analyse data efficiently",
)

task1 = Task(
    name="extract_data",
    description="Extract data from source files",
    expected_output="Extracted data in JSON format",
    agent=processor,
    retain_full_context=False,
)

task2 = Task(
    name="analyse_trends",
    description="Analyse trends across all previous data",
    expected_output="Comprehensive trend analysis",
    agent=processor,
    retain_full_context=True,
)

team = AgentTeam(agents=[processor], tasks=[task1, task2], process="sequential")
team.start()

Configuration Options

Option	Type	Default	Description
`retain_full_context`	`bool`	`False`	When `True`, pass all prior task outputs; when `False`, pass only the immediate predecessor’s output

Common Patterns

Pipeline pattern

Each step passes only its own output forward — ideal for large intermediate data:

pipeline_tasks = [
    Task(name="fetch", description="Fetch data", agent=fetcher, retain_full_context=False),
    Task(name="validate", description="Validate format", agent=validator, retain_full_context=False),
    Task(name="transform", description="Transform data", agent=transformer, retain_full_context=False),
    Task(name="store", description="Store in database", agent=storer, retain_full_context=False),
]

Checkpoint pattern

Periodically retain full context so a later task can see the full history:

checkpoint_interval = 5
tasks = []

for i in range(20):
    is_checkpoint = (i % checkpoint_interval == 0)
    tasks.append(Task(
        name=f"task_{i}",
        description=f"Process item {i}",
        agent=processor,
        retain_full_context=is_checkpoint,
    ))

Summarisation checkpoints

Insert a summary task that reads full context, then continue with minimal context:

summary_task = Task(
    name="summarise_progress",
    description="Summarise all previous outputs concisely",
    agent=summariser,
    retain_full_context=True,
)

next_task = Task(
    name="continue",
    description="Continue from the summary",
    agent=processor,
    retain_full_context=False,
)

Best Practices

Plan context flow upfront

Map which tasks need full history versus only the previous output before running long workflows. Data-loading steps usually need retain_full_context=False; final synthesis steps often need True.

Use checkpoints in long pipelines

For workflows with many steps, add periodic summary or checkpoint tasks with retain_full_context=True so downstream tasks can recover context without carrying every intermediate payload.

Avoid token limit errors

If later tasks hit context limits, set retain_full_context=False on early heavy tasks and pass summaries instead of raw data.

Combine with context compaction

Pair selective retention with Context Window Management for automatic overflow protection on agents.

Context Window Management

Automatic token management for agents

Workflow Validation

Validation loops with controlled context

Planning Mode

Fast Context

​How It Works

​Quick Start

​Configuration Options

​Common Patterns

​Pipeline pattern

​Checkpoint pattern

​Summarisation checkpoints

​Best Practices

​Related