Core Concepts
Key concepts you need to understand to work effectively with the framework.
Project Templates
A Project Template defines the structure and behavior of a workflow. It specifies what stages the project goes through and how content is generated.
A template is like a blueprint. The "Website" template defines how websites are built. The "Ebook" template defines how books are written.
Each template includes:
- Stages: The steps in the workflow
- Default Settings: Provider, model, parameters
- Workflow Model: The Python class that implements custom logic
Workflow Stages
Stages are the individual steps in a workflow. Each stage has a specific purpose and execution mode.
Execution Modes
| Mode | Description | Example |
|---|---|---|
template |
AI processes using a prompt template | Generate page content |
manual |
Requires human action | Review and approve |
auto |
Automatic system processing | Create child tasks |
Stage Properties
- Auto Advance: Move to next stage automatically when complete
- Require Approval: Wait for human approval before advancing
- AI Template: The prompt template to use for AI stages
AI Providers
AI Providers are the AI services that generate content. The framework supports multiple providers:
- OpenAI: GPT-4, GPT-3.5-turbo
- Anthropic: Claude 3 Opus, Sonnet, Haiku
- Custom: Any API-compatible service
You can configure multiple providers and assign different ones to different templates or stages based on cost, speed, or quality requirements.
Tasks
Tasks are the work units within a project. Each task represents a piece of content to be generated or reviewed.
Task States
pending → queued → processing → done
↓
failed| State | Description |
|---|---|
pending |
Waiting to be processed |
queued |
Added to processing queue |
processing |
AI is currently generating content |
done |
Successfully completed |
failed |
Error occurred during processing |
Approval Gates
Approval Gates are checkpoints where human review is required. They ensure quality control and allow users to guide the AI's direction.
Without approval gates, AI might generate content that doesn't match user expectations. Strategic checkpoints let users course-correct early, before significant resources are spent on wrong content.
Typical Approval Points
- After Discovery: Approve the overall plan/structure
- After Generation: Approve individual content pieces
- Before Publishing: Final review before going live
When approval is needed, users can:
- Approve: Accept and continue to next stage
- Request Changes: Provide feedback for AI to revise
- Reject: Cancel and start over