What are Members?
In Conductor’s musical metaphor, members are the musicians in your orchestra. Each member plays one instrument (role) and does it well. Members are the fundamental building blocks that execute specific tasks within an ensemble.Why “Members” instead of “Agents”?While members ARE agents that perform tasks, we use this term to:
- Reinforce the musical metaphor (members = musicians in the orchestra)
- Distinguish from traditional autonomous AI agents
- Emphasize single-responsibility principle (each member has one clear role)
- Create consistency with “ensemble” terminology
Think Members
AI reasoning with LLMs (GPT-4, Claude, Workers AI)
Function Members
JavaScript/TypeScript business logic
Data Members
Storage operations (KV, D1, R2, Vectorize)
API Members
HTTP requests to external services
Member Structure
Every member has two parts:1. Configuration (member.yaml)
Defines the member’s metadata, type, and configuration:
2. Implementation (index.ts)
The actual code that runs when the member executes:
Member Types Explained
Think Members
AI reasoning members that call LLM providers:- Natural language understanding
- Content generation
- Classification and extraction
- Reasoning and analysis
- OpenAI (GPT-4, GPT-4o, o1)
- Anthropic (Claude 3.5, Claude 3)
- Workers AI (Edge models)
- Groq (Ultra-fast inference)
- Custom providers
Function Members
JavaScript/TypeScript code execution:- Data transformation
- Business logic
- Calculations
- Formatting and validation
Data Members
Storage operations with Cloudflare primitives:get- Retrieve dataput- Store datadelete- Remove datalist- Query multiple items
- KV - Key-value cache (global, eventually consistent)
- D1 - SQLite database (serverless SQL)
- R2 - Object storage (S3-compatible)
- Vectorize - Vector database (embeddings)
API Members
HTTP requests to external services:- Automatic retry with exponential backoff
- Timeout handling
- Header interpolation
- JSON/text response parsing
Member Lifecycle
Execution Flow
- Load: Member configuration is loaded
- Cache Check: Check if cached response exists
- Execute: Run the member’s implementation
- Wrap: Format response with metadata
- Cache: Store result (if caching enabled)
- Return: Provide result to ensemble
Response Format
Every member returns a standardized response:Built-In Members
Conductor provides 6 production-ready built-in members:scrape - Web Scraping
scrape - Web Scraping
3-tier fallback strategy: Cloudflare Browser Rendering → Puppeteer → HTML parsing
validate - Quality Evaluation
validate - Quality Evaluation
4 evaluator types: rule-based, LLM judge, NLP, embedding similarity
rag - RAG System
rag - RAG System
Vectorize integration with semantic search
hitl - Human-in-the-Loop
hitl - Human-in-the-Loop
Approval workflows with notifications
fetch - HTTP Client
fetch - HTTP Client
Retry logic and timeout handling
queries - SQL Execution
queries - SQL Execution
Hyperdrive integration for databases
Member Best Practices
1. Single Responsibility
Each member should do one thing well:2. Explicit Dependencies
Declare what state you need:3. Enable Caching
Cache expensive operations:4. Use Schema Validation
Define input/output schemas:5. Handle Errors Gracefully
Return structured errors:Member Discovery
Members are discovered in this order:- Built-in members (
scrape,validate,rag,hitl,fetch,queries) - User-defined members (in your project’s
members/directory) - Versioned members (name@version via Edgit - planned)

