Blueprints
Visual workflow system that powers intelligent agent execution and complex automation
Visual Workflow Engine: Blueprints power every agent interaction - from simple conversations to complex multi-step automations through an intuitive node-based system.
What Are Blueprints?
Blueprints are the visual workflow system that powers how agents and tools execute tasks. Every conversation, tool call, and automation in Starships.ai runs through blueprints - making them the foundational execution engine of the platform.
- Visual Workflows: Node-based system for building complex logic without code
- Intelligent Execution: AI-driven decision making with conditional branching
- Tool Integration: Seamless connection to external APIs and services
- State Management: Persistent data flow across multi-step processes
How You'll Use Blueprints
Blueprints work behind the scenes in three key ways:
- Agent Conversations: Every chat follows a blueprint workflow automatically
- Template Browsing: See blueprint visualizations in marketplace template details
- Custom Tools: Daedalus creates blueprints when you describe automation needs
Universal Blueprint Architecture
Every blueprint follows the same core execution pattern, regardless of complexity:
Core Components
Every blueprint is built from these fundamental elements:
- Nodes: Individual processing steps that perform specific actions
- Edges: Connections that define execution flow and data passing
- State: Shared data that persists across node executions
- Data Flow: Simple expressions that connect node outputs to inputs
How Blueprints Execute
When a blueprint runs, the execution engine follows this process:
Initialize State
The workflow begins with the start node, which receives input data and initializes the workflow state.
Process Nodes
Each node executes in sequence, with data flowing along edges. Nodes can be:
- Processing (transform data, make API calls)
- Decision (branch based on conditions)
- Integration (call tools, external services)
Manage State
State persists across nodes, allowing complex data transformations and multi-step processes to maintain context.
Return Results
The workflow completes at an end node, returning structured results back to the calling agent or system.
Blueprint Data Flow
Blueprints use simple expressions to pass data between workflow steps:
// Access workflow state
'{{ state.userEmail }}';
// Get results from previous nodes
"{{ lastOutputOf('api_call') }}";
// Use input data
"{{ input.message }}";Common Patterns:
- State Variables:
{{ state.variableName }}- Access persistent workflow data - Node Results:
{{ lastOutputOf('nodeRef') }}- Get outputs from previous steps - Input Access:
{{ input.path }}- Extract data from workflow input
Node Reference Guide
Blueprints are built from 13 specialized node types, each designed for specific workflow functions:
| Node Type | Purpose | Key Parameters | Common Use Cases |
|---|---|---|---|
start | Workflow entry point | input schema | All workflows, input validation |
end | Workflow exit point | output data | Return results, workflow completion |
llm | AI model processing | model, prompt, temperature | Text analysis, reasoning, generation |
conditional | Branch execution flow | expression, conditions | Decision points, logic routing |
http_request | External API calls | url, method, headers, auth | API integrations, data fetching |
connector_action | Connector-backed API actions | connector, action, args, timeout | Managed integrations, auth-injected requests, schema-validated actions |
code | JavaScript execution | script, timeout | Data processing, calculations |
tool | Function execution | tool reference, parameters | Custom functions, tool calling |
transform | Data transformation | JSON patches, mappings | Format conversion, data shaping |
mutate | State modification | reducer function | Update workflow state, counters |
workflow | Nested blueprint execution | blueprint reference | Reusable workflows, composition |
agent | Complete agent execution | agent reference, context | Delegate to agents, conversations |
user_prompt | Human interaction | prompt text, input type | User input, approval workflows |
connector_action vs http_request
- Use
http_requestwhen you need a raw HTTP call to an absolutehttp(s)://...URL and you can provide everything needed in the node (method/headers/body), like public APIs, webhooks, or simple integrations. - Use
connector_actionwhen calling a Starships connector action (defined by connector templates) so the runtime can handle base URLs, auth/credentials injection, argument validation, and standardized request/response mapping—especially for integrations that should not embed secrets directly in blueprints.
Advanced Workflow Patterns
Blueprints support sophisticated workflow patterns for complex automation needs:
Sequential Processing Pattern
For straightforward workflows with ordered steps:
Use Cases: Data processing pipelines, API integrations, file operations
Conditional Branching Pattern
For workflows requiring different paths based on data or conditions:
Use Cases: User role routing, content type handling, error recovery strategies
Loop and Retry Pattern
For workflows requiring iteration or error recovery:
Use Cases: API retry logic, batch processing, polling operations, error recovery
Real-World Examples
Blueprints power real automation workflows in the marketplace:
- RAG Search & Chat: Multi-step vector search with AI-powered response generation
- Support Ticket Triage: Automated classification and routing based on content analysis
- Lead Enrichment: Sequential API calls to gather and consolidate customer data
- Content Moderation: Parallel processing with rule-based and AI-powered validation
- API Orchestrator: Complex multi-service coordination with error handling