Workflow History & Continuous Execution

Build workflows that reference previous runs across multiple executions using workflow history.

Before running the examples, create and activate a virtual environment:

uv pip install agno openai sqlalchemy
export OPENAI_API_KEY="your-api-key"
v2.1.4

Workflow History enables your Agno workflows to remember and reference previous conversations, transforming isolated executions into continuous, context-aware interactions.

Instead of starting fresh each time, with Workflow History you can:

  • Build on previous interactions - Reference the context of past interactions
  • Avoid repetitive questions - Avoid requesting previously provided information
  • Maintain context continuity - Create a conversational experience
  • Learn from patterns - Analyze historical data to make better decisions

This feature is different from add_history_to_context. It adds prior completed workflow input/output pairs to selected steps. These pairs are distinct from an individual agent's message history and from preceding step output in the current run.

The fragments below reuse this setup:

from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.workflow import Step, Workflow
from agno.workflow.types import StepInput, StepOutput

db = SqliteDb(db_file="tmp/history_workflow.db")
research_agent = Agent(name="Research", instructions="Summarize the supplied topic.")
analysis_agent = Agent(name="Analysis", instructions="Analyze the preceding summary.")
writing_agent = Agent(name="Writer", instructions="Write a short answer from the supplied context.")
content_agent = writing_agent
research_step = Step("Research", agent=research_agent)
analysis_step = Step("Analysis", agent=analysis_agent)
writing_step = Step("Writing", agent=writing_agent)

How It Works

When workflow history is enabled, the last three completed runs' input/final-output pairs are injected into agent/team step inputs by default:

<workflow_history_context>
[Workflow Run-1]
User input: Create content about AI in healthcare
Workflow output: # AI in Healthcare: Transforming Patient Care...

[Workflow Run-2]
User input: Make it more family-focused
Workflow output: # AI in Family Healthcare: A Parent's Guide...

</workflow_history_context>

Your current input goes here...

Along with this, in using Steps with custom functions, you can access this history in the following ways:

  1. As a formatted context string as shown above
  2. In a structured format as well for more control
[
    ("<workflow input from run 1>", "<workflow output from run 1>"),
    ("<workflow input from run 2>", "<workflow output from run 2>"),
]

A database is required to use Workflow history. Runs across different executions will be persisted there.

Example:

def custom_function(step_input: StepInput) -> StepOutput:
    # Option 1: Structured data for analysis
    history_tuples = step_input.get_workflow_history(num_runs=3)
    for user_input, workflow_output in history_tuples:
        ...  # Process each conversation turn

    # Option 2: Formatted context for agents
    context_string = step_input.get_workflow_history_context(num_runs=3)

    return StepOutput(content="Analysis complete")

You can use these helper functions to access the history:

  • step_input.get_workflow_history(num_runs=3)
  • step_input.get_workflow_history_context(num_runs=3)

Refer to StepInput reference for more details.

Control Levels

You can be specific about which Steps to add the history to:

Workflow-Level History

Add workflow history to all steps in the workflow:

workflow = Workflow(
    db=db,
    steps=[research_step, analysis_step, writing_step],
    add_workflow_history_to_steps=True  # All steps get history
)

Step-Level History

Add workflow history to specific steps only:

Step(
    name="Content Creator",
    agent=content_agent,
    add_workflow_history=True  # Only this step gets history
)

You can also put add_workflow_history=False to disable history for a specific step.

Precedence Logic

Step-level settings always take precedence over workflow-level settings:

workflow = Workflow(
    db=db,
    steps=[
        Step("Research", agent=research_agent),                              # None → inherits workflow setting
        Step("Analysis", agent=analysis_agent, add_workflow_history=False),  # False → overrides workflow
        Step("Writing", agent=writing_agent, add_workflow_history=True),     # True → overrides workflow
    ],
    add_workflow_history_to_steps=True  # Default for all steps
)

History Length Control

By default, each step receives the last 3 runs (num_history_runs=3). Keep this limit small to avoid bloating the LLM context window.

Set num_history_runs on each agent/team Step when you need a different window. In the current implementation, the Step default of three overrides a larger Workflow(num_history_runs=...) value.

workflow = Workflow(
    db=db,
    add_workflow_history_to_steps=True,
    steps=[
        Step("Research", agent=research_agent, num_history_runs=5),
        Step("Analysis", agent=analysis_agent, num_history_runs=3),
        Step("Writing", agent=writing_agent, num_history_runs=5),
    ],
)
workflow.print_response("Explain workflow history", session_id="history-demo")
workflow.print_response("Give a shorter explanation", session_id="history-demo")

For a custom function, the num_runs argument on StepInput's history helpers controls its window independently.

Developer Resources

Single Step Continuous Execution

Single step workflow with continuous execution and history awareness.

Workflow History for Steps

Add workflow history to all steps in the workflow.

Enable History for Specific Step

Enable workflow history for a specific step only.

Get History in Function

Access workflow history in custom functions for analysis.

Multi-Purpose CLI

Add workflow history to the steps of a multi-purpose CLI workflow.

Intent Routing

Route requests to specialist agents that share the same conversation history.