BindAI Workflow

bindai-workflow provides workflow orchestration for the BindAI framework.

It provides the abstractions needed to define, execute, schedule, persist, serialize, deploy, and manage workflows composed of different types of execution nodes.

Features

• Workflow definitions
• Workflow builder and configuration
• Workflow execution instances
• Workflow registries
• Workflow storage
• In-memory workflow storage
• Agent nodes
• Runnable nodes
• Conditional nodes
• Parallel execution
• Join nodes
• Loop nodes
• Sub-workflows
• Start and end nodes
• Human tasks
• Retry policies
• Timeout policies
• Workflow scheduling
• Compensation actions
• Workflow history
• Workflow repositories
• Serialization and deserialization
• Workflow deployment
• Workflow factories
• Workflow management services

Public API

The package exposes the main workflow components directly:

from bindai_workflow import (
    Workflow,
    WorkflowBuilder,
    WorkflowConfiguration,
    WorkflowRegistry,
    WorkflowInstance,
    WorkflowStore,
    MemoryWorkflowStore,
)

Additional workflow infrastructure includes:

from bindai_workflow import (
    WorkflowFactory,
    WorkflowManager,
    WorkflowNodeFactory,
    WorkflowSerializer,
    WorkflowDeserializer,
    WorkflowRepository,
    WorkflowDeployment,
    WorkflowServices,
)

Creating Workflows

WorkflowBuilder provides the builder API for constructing workflows.

from bindai_workflow import WorkflowBuilder

builder = WorkflowBuilder(...)

Workflow configuration is represented by WorkflowConfiguration.

from bindai_workflow import WorkflowConfiguration

The resulting Workflow represents the executable workflow definition.

Workflow Nodes

Workflows can be composed from different node types.

The package exposes:

Workflow
   │
   ├── StartNode
   ├── AgentNode
   ├── RunnableNode
   ├── ConditionNode
   ├── ParallelNode
   ├── JoinNode
   ├── LoopNode
   ├── SubWorkflowNode
   ├── HumanTask
   ├── CompensationNode
   └── EndNode

These nodes allow workflows to represent different execution patterns and control-flow structures.

Parallel Execution

ParallelNode provides parallel workflow execution, while JoinNode represents synchronization after parallel branches.

from bindai_workflow import (
    ParallelNode,
    JoinNode,
)

This makes it possible to model workflows where independent operations can execute concurrently before continuing to a later stage.

Conditions and Loops

Conditional and iterative execution are supported through:

from bindai_workflow import (
    ConditionNode,
    LoopNode,
)

WorkflowExpression is also available for workflow expressions:

from bindai_workflow import WorkflowExpression

Sub-Workflows

SubWorkflowNode allows a workflow to incorporate another workflow as part of its execution.

from bindai_workflow import SubWorkflowNode

This supports composing larger workflows from reusable workflow definitions.

Human Tasks

HumanTask provides a workflow task abstraction for human-involved execution.

from bindai_workflow import HumanTask

This allows workflow definitions to represent work that is not performed entirely automatically.

Retry and Timeout Policies

Workflow execution can use retry and timeout policies:

from bindai_workflow import (
    RetryPolicy,
    TimeoutPolicy,
)

These policies provide workflow-level execution controls without requiring every workflow node to implement its own retry or timeout behavior.

Workflow Scheduling

Workflows can be scheduled through WorkflowSchedule and WorkflowScheduler.

from bindai_workflow import (
    WorkflowSchedule,
    WorkflowScheduler,
)

The scheduler provides the workflow-level scheduling abstraction, while the schedule represents the scheduling configuration.

Workflow State and Instances

WorkflowInstance represents an execution instance of a workflow.

from bindai_workflow import WorkflowInstance

Workflow persistence is represented by WorkflowStore:

from bindai_workflow import WorkflowStore

BindAI also provides an in-memory implementation:

from bindai_workflow import MemoryWorkflowStore

This separation allows workflow execution state to be stored independently from the workflow definition.

Workflow History

Workflow execution history is represented by WorkflowHistory and can be persisted through a history store.

from bindai_workflow import (
    WorkflowHistory,
    MemoryHistoryStore,
)

MemoryHistoryStore provides an in-memory history implementation.

Serialization

Workflows can be serialized and deserialized through:

from bindai_workflow import (
    WorkflowSerializer,
    WorkflowDeserializer,
)

Factories are also available for constructing workflows:

from bindai_workflow import (
    WorkflowFactory,
    WorkflowNodeFactory,
)

Compensation

Compensation actions support workflows that need explicit recovery or compensating behavior.

from bindai_workflow import (
    CompensationAction,
    CompensationNode,
)

Workflow Management

Higher-level workflow management components include:

from bindai_workflow import (
    WorkflowManager,
    WorkflowServices,
)

These provide the management and service layer around workflow functionality.

Architecture

The workflow package separates workflow definitions from execution state and supporting infrastructure:

Workflow
   │
   ├── Nodes
   │     ├── Agent
   │     ├── Runnable
   │     ├── Condition
   │     ├── Parallel
   │     ├── Join
   │     ├── Loop
   │     ├── SubWorkflow
   │     ├── Human Task
   │     └── Compensation
   │
   ├── WorkflowInstance
   │
   ├── WorkflowStore
   │
   ├── WorkflowHistory
   │
   └── WorkflowScheduler

Supporting components provide serialization, deployment, repositories, factories, and workflow management.

BindAI Integration

bindai-workflow integrates with other BindAI packages to orchestrate executable components.

For example, workflow nodes can coordinate agents and runnable components while workflow state, scheduling, retry, timeout, and history concerns remain within the workflow layer.

Development

This package is part of the BindAI workspace.

From the repository root:

uv sync

Run the complete test suite:

uv run pytest

Run static type checking for this package:

uv run mypy packages/bindai-workflow

Build the package:

uv build --package bindai-workflow

Documentation

For complete BindAI framework documentation:

https://docs.bindai.dev

License

BindAI is released under the MIT License.