ombra 0.1.0

Simple AI workflow orchestration with time travel for development

Ombra - Simple AI Workflow Orchestration with Time Travel

Simple Python workflow orchestration with checkpointing and time travel for development.

Installation

For local development:

# From this directory
uv run examples/test_basic.py

# Or install locally in another project
uv add --editable /path/to/ombra

Quick Start

from ombra import Workflow

# Create a workflow
wf = Workflow(name="my_workflow")

# Decorate your functions
@wf.step()
def fetch_data(query: str) -> dict:
    return {"query": query, "results": ["item1", "item2"]}

@wf.step()
def process_data(data: dict) -> list:
    return [item.upper() for item in data["results"]]

# Run normally - checkpoints saved automatically
data = fetch_data("AI frameworks")
processed = process_data(data)

# List all checkpoints
wf.list_checkpoints()

# Time travel: Resume from any step
result = wf.resume_from("process_data")

# Resume with modified inputs
result = wf.resume_from(
    "process_data",
    modify_inputs={"data": {"query": "test", "results": ["new"]}}
)

Features

• Decorator-based API - Write normal Python functions
• Automatic checkpointing - Every step execution is saved
• Time travel - Resume from any step
• Modify inputs - Change inputs when resuming for debugging
• File-based storage - Simple pickle files, no external services
• Zero dependencies - Just Python stdlib

Checkpoints

Checkpoints are saved to .ombra_checkpoints/{workflow_name}/ directory. Each checkpoint includes:

• Step name and timestamp
• Input arguments (args and kwargs)
• Output values
• Metadata

Development Only

This is designed for development and debugging workflows, not production. The checkpoint system helps you:

• Debug complex workflows step by step
• Avoid re-running expensive operations
• Test different inputs at any step
• Understand workflow state at each stage

Next Steps

This is a minimal MVP. Future additions:

• Async support
• Automatic flow chaining
• DAG support for parallel execution
• Retry and error handling decorators
• SQLite storage option
• Better visualization of checkpoints