Summarize Node

AI/Generative

Summarize

Summarizes long text using an LLM with configurable strategies. Supports Map-Reduce (parallel, fast), Refine (sequential, coherent), Hierarchical (structure-aware), Hybrid (parallel + coherent), and Sliding Window (memory-efficient). Optional Chain of Density post-processing for optimal information density.

ai_llm_summarizellmLong running

Inputs13

Outputs4

Security exposure5/10

Packagellm

Ratings

Scores range from 0 to 10. Higher values mean more impact, exposure, or operational weight.

SecurityAttack surface and exposure impact.

5/10Medium

PrivacyPotential sensitivity of processed data.

5/10Medium

PerformanceRuntime or resource pressure.

4/10Medium

GovernancePolicy, audit, or compliance impact.

5/10Medium

ReliabilityOperational stability considerations.

3/10High

CostExternal or compute cost impact.

5/10Medium

Input Pins

Input

Execution

exec_in

Execution trigger

Model

Struct

model

Bit describing the provider/model to use for summarization

Schema enforced

Text

String

text

The long text to summarize (markdown supported)

Strategy

String

strategy

Summarization strategy: • Refine — sequential, best coherence, no parallelism • MapReduce — parallel chunking, fast, may lose cross-chunk context • Hierarchical — structure-aware tree, best for headed documents • Hybrid — MapReduce speed + Refine coherence polish • SlidingWindow — fixed memory buffer, best for very long documents

Default Refine

RefineMapReduceHierarchicalHybridSlidingWindow

Densification

String

densification

Post-processing to increase information density: • None — use the strategy output as-is • ChainOfDensity — iteratively compress to optimal density (~0.15 entities/token)

Default None

NoneChainOfDensity

Instructions

String

instructions

Optional focus instructions (e.g. 'focus on action items', 'use bullet points')

Prior Summary

String

prior_summary

Optional existing summary to build upon (used as initial context for Refine/Hybrid/SlidingWindow strategies)

Chunk Size

Integer

chunk_size

Maximum characters per chunk. Reduce for models with smaller context windows (default: 8000)

Default 8000

Chunk Overlap %

Integer

chunk_overlap

Overlap between adjacent chunks as percentage (0-50). Prevents information loss at boundaries (default: 10)

Default 10

Track Entities

Boolean

track_entities

Extract and track named entities across chunks to prevent information loss. Adds 2-3 extra LLM calls but improves factual preservation.

Default false

Concurrency

Integer

concurrency

Parallel requests for MapReduce/Hybrid strategies. 0 = unlimited, 1 = sequential (default: 4)

Default 4

Max Iterations

Integer

max_iterations

Safety limit on summarization passes. Each pass reduces total length (default: 5)

Default 5

Density Steps

Integer

density_steps

Number of Chain of Density refinement steps when densification is enabled (1-5, default: 3). Research shows step 3 is the human-preferred sweet spot.

Default 3

Output Pins

Output

Execution

exec_out

Fires once summarization is complete

Summary

String

summary

The final summarized text

Entities

String Array

entities

Tracked entities found in the document (only populated when Track Entities is enabled)

LLM Calls

Integer

llm_calls

Total number of LLM invocations used during summarization

Node Info

Internal name: ai_llm_summarize
Category: AI/Generative
Version: 4