--- title: "Exa Integration" description: "Give AI agents meaning-based web search through Exa. Agents search by neural embeddings or keywords, extract full page contents and summaries, find pages similar to a known URL, get cited answers to natural language questions, and run autonomous multi-source research tasks." url: https://flowrunner.ai/integrations/exa-ai date_modified: 2026-08-01T02:40:32-07:00 --- # Exa [AI](https://flowrunner.ai/integrations/category/ai-llms) Give AI agents meaning-based web search through Exa. Agents search by neural embeddings or keywords, extract full page contents and summaries, find pages similar to a known URL, get cited answers to natural language questions, and run autonomous multi-source research tasks. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 7 actions · API key · available [Exa website](https://exa.ai/) · [Platform Documentation](https://docs.exa.ai/) · Capability data verified 2026-07-27 1. The Monday competitive-scan schedule fires 2. Search runs meaning-based queries over the past week with domain and date filters, extracting summaries inline 3. Find Similar Links expands coverage outward from each newly discovered company page 4. The agent dedupes results against the running log so only genuinely new sources move forward 5. Create Research Task turns the week's finds into a cited report with a structured output schema 6. The strategy lead reads the report, opens the citations behind any surprising claim, and marks what becomes company knowledge 7. The approved brief publishes to \[Notion\](/integrations/notion) and the digest posts to \[Slack\](/integrations/slack) ## What This Integration Enables Exa is web search built for machines to consume rather than humans to click: results arrive as parsed text, highlights, and summaries, ranked by meaning through neural embeddings instead of term overlap. That makes it the discovery layer for agent workflows. FlowRunner agents search by meaning or keyword with domain, date, and category filters, pull full page contents without a scraping stack, expand from a known URL to its semantic neighbors, get cited answers to direct questions, and launch autonomous research tasks that return structured, machine-readable reports. What the machine finds, a [human-in-the-loop](https://flowrunner.ai/concepts/human-in-the-loop) reviewer still judges before it becomes something the company believes. - Discover pages by meaning with neural search, or fall back to keyword and auto modes - Extract full text, highlights, and AI summaries inline, with live crawling for fresh pages - Expand from any URL to semantically similar sources, competitors, and alternatives - Run multi-source research tasks that return cited reports, optionally as structured JSON ### Without FlowRunner **Keyword search misses meaning**: A competitor's launch phrased in their words never matches your saved queries **Research is a browser marathon**: Someone loses an afternoon to forty tabs and pastes fragments into a doc **Claims arrive orphaned**: A "fact" circulates internally and nobody can trace where it came from ### With FlowRunner **Discovery works by meaning**: Neural search finds the pages that match the idea, whatever words they used **Research runs as a task**: An autonomous run explores sources for minutes and returns a report, not a link pile **Every claim carries its source**: Answers and reports come with citations a reviewer can actually open ## Use Case Scenarios ### The pre-call brief that builds itself A discovery call lands on the calendar via [Calendly](https://flowrunner.ai/integrations/calendly). The agent runs Answer with a direct question about what the prospect's company does and who they serve, then Get Contents on the company site and recent coverage for depth. Find Similar Links surfaces the competitors the prospect is most likely comparing against. The assembled brief, with every claim linked to its source, lands in the rep's [Slack](https://flowrunner.ai/integrations/slack) an hour before the call. The rep walks in informed and can verify anything that sounds off in one click, which matters more than the brief itself: a summary you can check is research, and a summary you cannot is a rumor. ### A reading list that curates itself The team tracks a technical space where the best material rarely uses predictable words. On a schedule, the agent runs Search in neural mode with a category focus on research papers and recent-date filters, then logs each new result's URL and summary to [Google Sheets](https://flowrunner.ai/integrations/google-sheets), skipping anything already captured. The sheet becomes a monitored reading list with the noise already filtered, and the weekly additions post to the team channel where a human picks what deserves a deep read. The list improves as the queries do, and the queries are versioned in the flow. ### Vendor diligence with a structured verdict Before a contract, operations wants the same questions answered about every vendor: funding, security posture, customer complaints, pricing signals. The agent calls Create Research Task with instructions and a JSON Schema for the output, polls Get Research Task until completion, and writes the structured result into the evaluation matrix. Because the schema is fixed, vendors become comparable rows rather than incomparable essays, and the citations ride along so the reviewer who signs the recommendation can check the load-bearing claims first. ## Human-in-Loop Highlight Answer and Create Research Task return synthesized prose that reads with total confidence, and the citations attached are exactly why a human gate belongs here: a citation proves a source was read, not that the sentence in front of you says what the source says. So no Exa-generated answer or research report enters the knowledge base, a customer-facing document, or a purchase recommendation on the agent's authority. The reviewer opens the citations behind the claims that would change a decision, confirms the sources actually carry them, and approves or sends it back. Research tasks also run for minutes across many sources and bill accordingly, so recurring research flows get their scope confirmed by a person before the schedule multiplies them. Agent processes routinely Detects exception requiring judgment Clear match Continues automatically Ambiguous Routes to human via preferred channel Human decides Agent resumes with decision ## Agent Capabilities 7 actions ### Search 2 - **Search** Runs an Exa search with a chosen mode: Neural for meaning-based discovery, Keyword for classic matching, Auto to let Exa pick per query, Fast for lower latency. Supports domain include and exclude lists, published-date ranges, phrase filters, and category focus, with optional inline extraction of full text, highlights, and summaries. - **Find Similar Links** Finds pages semantically similar to a given URL, with the same filters as Search and optional exclusion of the source's own domain. The expansion move for competitor discovery and related-source hunting, with inline contents available. ### Contents 1 - **Get Contents** Retrieves and parses specific pages by URL or prior result id, returning any combination of full text, highlights, and an AI summary. Livecrawl forces a fresh crawl for fast-changing pages, and Subpages pulls linked pages along. ### Answer 1 - **Answer** Asks a natural-language question and returns a synthesized answer grounded in live web search, with the supporting citations and optionally each source's full text. One call, one complete response. ### Research 3 - **Create Research Task** Starts an asynchronous research run that explores many sources, reasons over them, and produces a cited report from free-form instructions. Returns a task id immediately, and an optional JSON Schema turns the output into structured, machine-readable results. - **Get Research Task** Polls a research task's status and, on completion, returns the synthesized output with its gathered citations. - **List Research Tasks** Lists the account's research tasks, most recent first, with statuses and cursor pagination. The audit view over what has been asked and answered. ## Frequently Asked Questions ### What can FlowRunner do with Exa? FlowRunner agents can run Search, Get Contents, and Find Similar Links in Exa, plus 4 more actions. ### Does connecting Exa to FlowRunner require OAuth? No. Exa connects to FlowRunner with an API key, no OAuth flow required. ### Can Exa trigger a FlowRunner workflow automatically? Exa doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at Exa?** This integration exposes Exa 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/exa-ai. Site index: https://flowrunner.ai/llms.txt