When to Build a Plugin

​

Quick Decision Tree

​

Plugins: Custom Compute Logic

​

When to Use Plugins

​

Example: Text Analysis Plugin

You want agents to analyze sentiment in user messages:
1. Implement sentiment analysis algorithm
2. Compile to WebAssembly
3. Deploy as plugin
4. Agents call: my_sentiment_analyzer
5. Result: Custom logic runs safely in sandbox

​

Advantages

• Complete control - Implement exactly what you need
• Isolated execution - Runs in secure sandbox
• No external dependencies - Pure compute
• Low latency - In-process execution
• Versioning - Easy to iterate and rollback

​

Disadvantages

• Development overhead - Must write and test code
• Language lock-in - Supported languages only (Rust, Python, TypeScript, Go)
• Maintenance burden - Responsible for updates
• Testing complexity - Need test environments

​

Connectors: External Services

​

When to Use Connectors

​

Example: Airtable Integration

You want agents to read/write Airtable bases:
1. Create REST connector pointing to Airtable API
2. Define operations (GET /bases/{id}, PATCH /records)
3. Agents call: airtable_read_records
4. Connector handles HTTP, authentication, response parsing

​

Advantages

• No code required - Configure via UI
• Built-in auth - OAuth, API keys, bearer tokens supported
• API abstraction - Hide complexity behind simple tools
• Ecosystem reuse - Connect to 1000s of existing APIs
• Dynamic updates - Update without redeployment

​

Disadvantages

• External dependency - Relies on third-party API availability
• Latency - Network round-trip required
• Rate limits - Subject to API quotas
• Less control - Limited customization beyond parameter mapping

​

Built-ins: Standard Features

​

When Built-in vs Plugin

• Multiple users need this feature
• It’s a fundamental capability (like files or code execution)
• You want Noorle to maintain and support it

• It’s specific to your use case
• You need custom behavior or logic
• You want ownership and control

​

Example: Should I Build or Use Existing?

Scenario: "I need to extract entities from text"

Option 1: Plugin
- Implement NER algorithm in Rust/Python
- Full control, can customize for domain
- Only you maintain it
- Ideal if: You have proprietary ML models or special domain

Option 2: Connector
- Use spaCy or similar NLP APIs
- No code needed
- Third-party maintains it
- Ideal if: Existing service meets your needs

Option 3: Built-in (Code Runner)
- Execute spaCy/transformers in Python builtin
- Quick and simple
- Limited to available libraries
- Ideal if: Standard libraries are sufficient

​

Comparison Matrix

​

Common Use Cases

​

”I want to add data processing”

• Custom algorithms, transformations
• Offline processing

​

”I want to integrate with Salesforce”

• REST/OAuth integration
• No code needed

​

”I want agents to run Python code”

• Already available
• Execute arbitrary Python

​

”I want custom Salesforce logic”

• Complex orchestration
• Custom business rules
• Call Salesforce via HTTP within plugin

​

”I want web search integration”

• Already available
• No configuration needed

​

Decision Checklist

• Does a similar builtin already exist?
• Is there a public API I could use via connector?
• Do I need custom compute logic?
• Will this logic be reused across my organization?
• Can I maintain this code long-term?
• Do I need to keep the algorithm private/isolated?
• Is latency critical (plugins are faster)?
• Do I have the technical resources to develop it?

​

Getting Help

• Questions about capabilities? Check the capabilities list
• Want to see available connectors? Browse the connector overview
• Need custom code execution? Check Code Runner builtin
• Still unsure? Reach out to the Noorle community

​

Next Steps

• Building a plugin? Read Project Structure
• Building a connector? See REST connector guide
• Understanding plugins? Read Platform Overview