--- title: "AWS Comprehend Integration" description: "Run Amazon Comprehend natural-language processing on any text in a flow: sentiment, entities, key phrases, dominant language, and syntax, with native AWS Signature V4 signing." url: https://flowrunner.ai/integrations/aws-comprehend date_modified: 2026-08-07T17:20:23-07:00 --- # AWS Comprehend [AI](https://flowrunner.ai/integrations/category/ai-llms) Run Amazon Comprehend natural-language processing on any text in a flow: sentiment, entities, key phrases, dominant language, and syntax, with native AWS Signature V4 signing. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 9 actions · API key · available [AWS Comprehend website](https://aws.amazon.com/comprehend/) · [Platform Documentation](https://docs.aws.amazon.com/comprehend/) · Capability data verified 2026-07-14 1. New customer review lands in the feedback system 2. Agent runs Detect Sentiment and Detect Entities on the review text 3. Agent tags the review by sentiment and named product 4. Agent files the tagged review in the analytics store 5. Product channel gets a rollup of negative reviews by feature 6. Strongly negative reviews naming a person or safety issue route to a manager ## What This Integration Enables Comprehend turns unstructured text into structured signals. It scores sentiment, pulls out entities and key phrases, detects the dominant language, and returns syntax, all synchronously and fast enough to run on every record as it arrives. That makes it a natural first pass on any stream of free text: reviews, tickets, survey responses, inbound messages. The output is only useful if something acts on it. An orchestration layer takes the sentiment score and the entities and decides what happens next: file it, route it, or escalate it. It pulls a person in when the signal is strong enough to warrant one. FlowRunner is built for that layer, so text analysis becomes a routing decision rather than a dashboard nobody checks. ### Without FlowRunner **Feedback read by hand**: Someone reads every review to gauge tone and topic **No structure on text**: Free-text comments sit unclassified and unsearchable **Escalations missed**: Angry or urgent messages get buried in the volume ### With FlowRunner **Sentiment tagged automatically**: Every review scored and categorized as it arrives **Text made structured**: Entities and key phrases turn comments into filterable data **Escalations surfaced**: Strongly negative signals get flagged for a person immediately ## Use Case Scenarios ### Review Sentiment Routing Customer reviews arrive continuously. The agent runs Detect Sentiment and Detect Entities on each one, tags it by tone and by the product named, and files it in the analytics store. The product team gets a rollup of negative reviews grouped by feature instead of reading the full stream. The analysis that used to require a person skimming every review runs automatically as they land. ### Multilingual Ticket Sorting Support tickets arrive in several languages. The agent runs Detect Dominant Language on each ticket and routes it to an agent who speaks it, then extracts key phrases so the receiving agent sees the gist before opening it. Tickets reach the right person in the right language without a manual sorting step. ### Survey Response Analysis Open-ended survey responses are hard to act on at volume. The agent runs sentiment and key-phrase detection across every response, clusters the themes, and produces a summary of what people are asking for. Responses that score strongly negative and mention a specific risk are pulled out and routed to the owning team rather than folded into an aggregate. ## Human-in-Loop Highlight Sentiment analysis is good at scoring tone and bad at judging what a message deserves. When Comprehend scores a message as strongly negative or flags text that names a person or a safety concern, FlowRunner routes it through a [human-in-the-loop](https://flowrunner.ai/concepts/human-in-the-loop) step rather than filing it with the rest. The agent pauses, assembles the message and its scores, and sends it to a manager via Slack. A person decides whether it is a routine complaint or something that needs a real response. The model catches the signal; a human weighs it. Agent processes routinely Detects exception requiring judgment Clear match Continues automatically Ambiguous Routes to human via Slack Human decides Agent resumes with decision ## Agent Capabilities 9 actions ### Sentiment 3 - **Detect Sentiment** Analyzes a single document and returns the prevailing sentiment (POSITIVE, NEGATIVE, NEUTRAL, or MIXED) along with confidence scores for each sentiment class. - **Detect Targeted Sentiment** Performs entity-level (targeted) sentiment analysis on a single document, returning each identified entity along with its mentions and the sentiment expressed toward it. - **Batch Detect Sentiment** Analyzes up to 25 documents in a single request and returns the prevailing sentiment and confidence scores for each. ### Entities 3 - **Detect Entities** Identifies named entities (people, places, organizations, dates, quantities, and more) in a single document, returning each entity's type, text, confidence score, and character offsets. - **Detect PII Entities** Locates personally identifiable information (PII) such as names, addresses, emails, phone numbers, and account identifiers in a single document, returning each entity's type, confidence score, and character offsets. - **Batch Detect Entities** Identifies named entities in up to 25 documents in a single request. Returns a resultList (successful documents, each keyed by its input Index with its detected entities) and an errorList (documents that failed). ### Key Phrases 1 - **Detect Key Phrases** Extracts the key noun phrases from a single document, returning each phrase's text, confidence score, and character offsets. ### Language 1 - **Detect Dominant Language** Determines the dominant language of a single document, returning candidate language codes ranked by confidence score. ### Syntax 1 - **Detect Syntax** Performs part-of-speech tagging on a single document, returning each token's text, part-of-speech tag, confidence score, and character offsets. ## Frequently Asked Questions ### What can FlowRunner do with AWS Comprehend? FlowRunner agents can run Detect Sentiment, Detect Entities, and Detect Key Phrases in AWS Comprehend, plus 6 more actions. ### Does connecting AWS Comprehend to FlowRunner require OAuth? No. AWS Comprehend connects to FlowRunner with an API key, no OAuth flow required. ### Can AWS Comprehend trigger a FlowRunner workflow automatically? AWS Comprehend doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at AWS Comprehend?** This integration exposes AWS Comprehend to AI agents on every FlowRunner plan, including through MCP, at no cost to you. [See what FlowRunner offers integration partners](https://flowrunner.ai/integrations/partners), including how to keep this page current. --- Markdown version of https://flowrunner.ai/integrations/aws-comprehend. Site index: https://flowrunner.ai/llms.txt