
Oakie is a specialized system for building stateful, multi-step, long-running automations that require high-level accuracy. It includes the following components, each contributing to its robust automation capabilities:
- Oakie Interface: A no-code interface for creating and running automations with response verification, provenance tracking, and feedback tools.
- Oakie Graph: A cyclic, event-based, stateful graph designed to execute automations and workflow orchestrations reliably.
- Automation Orchestrator: Coordinates multi-step workflows with document processing, human-in-the-loop support, and API integration.
- Workflow Orchestrator: Oversees agentic workflows, involving multiple LLMs, tools, retries, and task coordination for complex processes.
- Document Ingestion: A library for ingesting, processing, and querying documents, with support for diverse formats and efficient data extraction.
- Adaptive Learning: Enables agents to improve over time using user feedback, statistical evaluations, and corrective adjustments.
Through the Oakie Interface, users can ingest documents and build, test, and interact with automated workflows. The execution of these workflows is managed by the Automation Orchestrator, while agentic tasks within the workflows are coordinated by the Workflow Orchestrator. Both orchestrators are built on the Oakie Graph, providing a robust foundation for complex automation processes. The system's integrated approach ensures reliable execution of workflows while maintaining state management and enabling continuous improvement through feedback and adaptations.
Oakie Interface

The Oakie Interface is a no-code platform designed for users to build, compile, and execute automations without requiring programming expertise.
- Side-by-Side Verification: Enables users to verify responses easily by comparing results side by side.
- Full Provenance: Tracks and displays attribution for all steps, ensuring transparency and traceability.
- Feedback and Editing: Allows users to provide feedback and edit responses, which feeds into adaptive learning mechanisms.
Oakie Graph

The Oakie Graph is a cyclic, event-based, stateful graph that forms the foundation for running automations and workflow orchestrators.
- Stateful Execution: Ensures workflows maintain state across events, enabling robust and adaptive automation.
- Event-Based Design: Processes are triggered and managed based on real-time events, enhancing responsiveness and efficiency.
Automation Orchestrator

The Automation Orchestrator manages user-defined automations by coordinating multi-step, agentic workflows.
- Human-in-the-Loop: Supports human intervention in workflows where manual decision-making or adjustments are necessary.
- Document Processing: Integrates capabilities for uploading and processing various document types within automation workflows.
- External API Integration: Provides an API interface for external tools to trigger and interact with automations.
Workflows Orchestrator

The Workflows Orchestrator handles agentic workflows as defined by the Automation Orchestrator. Each workflow step may involve multiple LLMs, tools, and retries.
- Multi-Agent Integration: Orchestrates tasks involving various agents and tools for complex workflows.
- Retry Mechanism: Ensures robustness by handling errors and retries effectively.
- Tool and Model Support: Integrates with a range of tools and large language models for optimized execution.
Document Ingestion

The Document Ingestion library enables efficient processing and querying of diverse document formats, supporting automation workflows that rely on document analysis.
- Format Compatibility: Handles ingestion of PDFs, HTML pages, Word documents, and more.
- Text and Data Extraction: Employs OCR and NLP techniques for extracting actionable data from documents.
- Query Optimization: Indexes and processes documents for fast and accurate semantic querying.
Adaptive Learning
The Adaptive Learning mechanism ensures Oakie’s workflows and agents continuously improve over time, based on user feedback and performance data.
- User Feedback Integration: Incorporates feedback from users to refine and optimize responses and workflows.
- Answer Evaluation: leveraging statistical analysis of multiple answers and integrating user feedback to assess the accuracy of responses
- Auto Correction: Utilizes feedback from the previous step to refine or test alternative approaches to answering a question.
Conclusion
Oakie’s architecture is purpose-built to manage complex, document-intensive automations with precision and adaptability. With its modular components, event-driven design, and focus on continuous improvement, Oakie represents a comprehensive solution for long-running automation scenarios. This overview provides insights into the technical foundation that powers Oakie, enabling it to deliver robust and reliable automation.