Table of Contents

Enum ChainType

Namespace
AiDotNet.Enums
Assembly
AiDotNet.dll

Represents different types of chains for composing language model operations.

public enum ChainType

Fields

Conditional = 1

Conditional chain with branching logic based on intermediate results.

For Beginners: Conditional chains make decisions and take different paths based on results.

Like a choose-your-own-adventure book:

  • Evaluate a condition
  • If true, go down path A
  • If false, go down path B
  • Different paths lead to different outcomes

Example: Input: Customer email

Step 1: Classify email type → If "complaint":

  • Extract issue
  • Check severity
  • Route to appropriate team → If "question":
  • Search knowledge base
  • Generate answer → If "feedback":
  • Categorize feedback
  • Log to database

The chain adapts based on what it encounters.

Use this when:

  • Different inputs require different processing
  • You need to make decisions based on intermediate results
  • Workflows have multiple possible paths
Loop = 2

Looping chain that repeats operations until a condition is met.

For Beginners: Looping chains repeat operations until they achieve a goal or meet a condition.

Like a video game character searching for a key:

  • Check current room for key
  • If found: Stop searching
  • If not found: Move to next room and repeat
  • Continue until key is found or all rooms are checked

Example - Research assistant: Goal: "Find 5 peer-reviewed articles about climate change"

Loop iteration 1: Search → Found 2 articles → Continue Loop iteration 2: Refine search → Found 1 more (total 3) → Continue Loop iteration 3: Broaden search → Found 2 more (total 5) → Stop

Result: 5 articles found

Loop with refinement: Goal: "Generate a summary that's exactly 100 words"

Attempt 1: Generated 85 words → Too short → Try again with "expand" instruction Attempt 2: Generated 112 words → Too long → Try again with "condense" instruction Attempt 3: Generated 98 words → Close enough → Accept

Use this when:

  • You need to refine results iteratively
  • The exact number of steps isn't known in advance
  • Quality improves with iteration
  • You need to meet specific criteria
MapReduce = 4

Map-reduce chain that processes collections by mapping operations and reducing results.

For Beginners: Map-reduce chains process large collections by working on items individually, then combining results.

Two phases:

  1. Map: Apply the same operation to each item in a collection
  2. Reduce: Combine all the results into a single output

Think of it like grading exams:

  • Map phase: Grade each student's exam individually (can be done in parallel)
  • Reduce phase: Calculate class average and statistics from all grades

Example - Analyze multiple documents: Input: [Document1, Document2, Document3, Document4, Document5]

Map phase (apply to each document):

  • Document1 → Extract key points → ["Point A", "Point B"]
  • Document2 → Extract key points → ["Point C", "Point D"]
  • Document3 → Extract key points → ["Point E"]
  • Document4 → Extract key points → ["Point F", "Point G"]
  • Document5 → Extract key points → ["Point H"]

Reduce phase (combine all results): All points: ["Point A", "Point B", "Point C", ..., "Point H"] → Identify common themes → Rank by importance → Generate final summary

Benefits:

  • Handles large collections efficiently
  • Map phase can be parallelized
  • Scales well with data size
  • Clean separation of processing and aggregation

Use this when:

  • Processing multiple similar items
  • Each item needs the same analysis
  • Final result combines all individual analyses
  • Working with large document collections
Parallel = 3

Parallel chain that executes multiple independent operations simultaneously.

For Beginners: Parallel chains run multiple operations at the same time, then combine results.

Like a team working on a project:

  • Different team members work on different tasks simultaneously
  • Each person completes their part independently
  • At the end, all parts are combined into the final product

Example - Comprehensive product analysis: Input: Product URL

Parallel operations (all at once):

  • Branch 1: Extract product specs
  • Branch 2: Analyze customer reviews
  • Branch 3: Find competitor prices
  • Branch 4: Check availability

Combine results: "Product X costs $99, has 4.5-star rating, is in stock, and is cheaper than competitors"

Benefits:

  • Much faster than sequential processing
  • Each branch is independent
  • Results are combined at the end

Use this when:

  • Operations don't depend on each other
  • Speed is important
  • You need multiple perspectives on the same input
  • Tasks can be done independently
Router = 5

Router chain that directs inputs to specialized chains based on content.

For Beginners: Router chains analyze input and send it to the best specialized chain for that type.

Like a receptionist at a hospital:

  • Patient arrives with symptoms
  • Receptionist determines the issue
  • Patient is directed to the right specialist
  • Specialist provides appropriate care

Example - Customer support system: Input: Customer message

Router analysis:

  • If about billing → Route to BillingChain
  • If about technical issue → Route to TechnicalSupportChain
  • If about product info → Route to ProductInfoChain
  • If about returns → Route to ReturnsChain

Each specialized chain is optimized for its specific task:

  • BillingChain: Knows about invoices, payments, refunds
  • TechnicalSupportChain: Knows about troubleshooting, diagnostics
  • ProductInfoChain: Knows about features, specifications
  • ReturnsChain: Knows about return policies, shipping

Benefits:

  • Each chain can be optimized for its specialty
  • Easier to maintain than one giant chain
  • Better results because of specialization
  • Can add new specialized chains easily

Use this when:

  • Inputs fall into distinct categories
  • Different categories need different processing
  • You want specialized handling for each type
  • One-size-fits-all approach doesn't work well
Sequential = 0

Simple sequential chain where output of each step feeds into the next.

For Beginners: Sequential chains run steps one after another in order.

Like a recipe:

  • Step 1: Gather ingredients
  • Step 2: Mix ingredients
  • Step 3: Bake mixture
  • Step 4: Serve result

Each step must complete before the next begins.

Example: Input: "Analyze customer feedback about our product" Step 1: Extract feedback → ["Great product!", "Too expensive", "Love it"] Step 2: Classify sentiment → [Positive, Negative, Positive] Step 3: Generate summary → "Overall positive (67%) with price concerns" Output: Summary report

Use this when:

  • Steps must happen in a specific order
  • Each step depends on the previous step's output
  • The workflow is straightforward and linear

Remarks

For Beginners: Chains connect multiple language model operations together, like building blocks.

Think of chains like assembly lines:

  • Each step does a specific task
  • Output from one step becomes input for the next
  • Complex workflows are built from simple components
  • The final result is the combination of all steps

For example, a research assistant might:

  1. Search for relevant documents (Step 1)
  2. Summarize each document (Step 2)
  3. Combine summaries into a final report (Step 3)

Chains make this easy to build, test, and modify.

Edit this page