{"slug": "which-search-api-lets-me-store-results-and-use-them-with-an-llm", "title": "Which search API lets me store results and use them with an LLM?", "summary": "A developer's review of published terms for eight search APIs found that Perplexity's Search API gives customers ownership of all output with no storage limits and no training on their data, while Brave permits only transient storage unless customers buy a custom plan, Parallel bars reusing one query's output for other end customers, and Exa bars copying anything obtained from the service without written permission. Tavily's terms state that it and its AI providers may use, process, analyze and retain customer input and outputs to train models, and SerpApi retains search data for 31 days unless ZeroTrace mode is enabled. The review also found that Brave's September 1, 2026 terms dropped an earlier AI inference restriction.", "body_md": "The Perplexity Search API has the clearest terms for this: the customer \"owns all Output\", Perplexity will not train on it, and no clause limits how long you keep it. Tavily, SerpApi and Serper also have no clause that limits storage or LLM use. Brave allows only transient storage unless you buy a custom plan, Parallel bars reusing one query's output for other users, and Exa bars copying anything you get from the service without its written permission. The search tools built into OpenAI, Claude, Gemini and Azure add display rules, and Gemini's are the strictest.\n\nThis is a reading of published terms. It is not legal advice. Each clause is quoted and linked, so you can check it against your own use.\n\nRead on November 16, 2026, from each vendor's current terms page.\n\n| API | Store or cache results | Train a model on results | Display or attribution | Terms | \n|---|---|---|---|---|\n| Brave Search API | Only \"transient storage required for operation of Customer Applications\" (§3(b)(i)). Storage rights need a custom plan | Forbidden: \"create, evaluate, train, re-train, fine-tune, benchmark\" (§3(b)(xiii)) | Optional \"POWERED BY BRAVE\" plus logo (§4(d)) | [1 Sep 2026](https://api-dashboard.search.brave.com/terms-of-service) | \n| Exa | §4.2(a) forbids copying, distributing or making derivative works from \"any information contained on, or obtained from or through, the Services\", unless Exa permits it in writing | No separate clause. Using Output \"to develop any competitive product\" is forbidden (§4.2(f)) | No clause found | [No date on the PDF](https://exa.ai/terms) | \n| Parallel | Output from one query is \"primarily for the use of one End Customer only, and shall not be copied, cached, stored, or made available to other End Customers\" (§2(b)) | Forbidden: \"create synthetic training data\" (§2(c)(vi)) | No clause found | [Customer Terms](https://parallel.ai/customer-terms) | \n| Tavily | No clause found | No clause on your side | No clause found | [May 4, 2026](https://www.tavily.com/terms) | \n| Perplexity Search API | No clause found. Output is your Confidential Information (§5.1) | No clause on your side | Display \"solely within the Customer Applications\" (§2.1) | [Jan 23, 2026](https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service) | \n| SerpApi | No clause found | No clause found | No clause found | [Aug 27, 2026](https://serpapi.com/legal) | \n| Serper | No clause found | No clause found | Do not \"misrepresent the ownership or the source\" | [May 29, 2024](https://serper.dev/terms) | \n| cloro | No clause found. You keep \"all rights\" in content you retrieve (§6.2) | Only lawful AI use (§7.3) | No clause found | [19 July 2026](https://cloro.dev/terms/) | \n\nThree details change the picture.\n\n**The data flows both ways.** Tavily §6.5 says Tavily and its AI providers \"may use, process, analyze, and retain Customer Input submitted to the AI Functionality and Outputs\" to train models. That covers Tavily's AI research features. Parallel §4(b) says it \"may use Customer IP to train and improve\" its models, and Customer IP includes your outputs. Perplexity §2.3.3 says the opposite: it \"shall not use (or authorize third parties to use) Customer Content to train\". SerpApi keeps search data for 31 days unless you turn on its ZeroTrace mode.\n\n**Parallel also blocks answer caches.** §2(c)(xi) forbids \"making available Customer Outputs from previous queries for future uses\". A shared cache that serves one user's results to the next user is the pattern it describes.\n\n**Brave's old \"AI inference\" rule is gone.** In early 2026 a [Hacker News post](https://news.ycombinator.com/item?id=46822822) said the terms of one Brave plan \"prohibit using responses for AI inference\", and a reply said Brave sold a separate AI plan. The terms of September 1, 2026 have no such clause. §3(a)(ii) lets you \"use the Search Results with the Customer Applications\", and the storage and training limits above still apply.\n\nThe model vendors' search tools are stricter, because the result is shown to a person.\n\n| Tool | Rule | \n|---|---|\n| OpenAI `web_search` | \"When your interface displays web results or information from them, you must show inline citations as prominent, clickable links.\" ( [docs](https://developers.openai.com/api/docs/guides/tools-web-search) ) | \n| Claude web search | \"When displaying API outputs directly to end users, citations must be included to the original source.\" Result text comes back as `encrypted_content` . ([docs](https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool) ) | \n| Gemini Grounding with Google Search | \"You will not ... cache, frame, syndicate, resell, analyze, train on, or otherwise learn from Grounded Results or Search Suggestions.\" You may store the text for up to two years for display tuning, chat history, or a follow-up prompt. ( [terms](https://aistudio.google.com/docs/terms#grounding-with-google-search) ) | \n| Grounding with Bing Search | \"Developers and end users don't have access to raw content returned from Grounding with Bing Search.\" Website URLs and the Bing query URL \"must be retained and displayed in the exact form provided by Microsoft.\" ( [docs](https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/bing-grounding) ) | \n\nGemini's terms also forbid using links \"to build an index\". If you want a dataset of results, a model-native grounding tool is the wrong source.\n\nThere is a pattern. APIs that run their own index say they own the results and set limits: Brave §9(a)(ii) says Brave is \"the sole and exclusive owner\" of the Search Results. APIs that return Google's results page set few limits on output, because the content is not theirs to license. The risk then sits with the source.\n\nGoogle sued SerpApi in December 2025. Google said SerpApi \"deceptively takes content that Google licenses from others\" ([Google](https://blog.google/technology/safety-security/serpapi-lawsuit/)). The court granted SerpApi's motion to dismiss on July 20, 2026 ([SerpApi](https://serpapi.com/blog/google-v-serpapi-the-court-granted-our-motion-to-dismiss/)), and then dismissed the amended complaint with leave to amend by November 30, 2026 ([Search Engine Roundtable](https://seroundtable.com/google-loses-again-against-serpapi-42508.html)). The court did not rule that the scraping is lawful.\n\nTwo vendors handle this risk differently:\n\n| What you want to do | Terms that allow it without a special deal | \n|---|---|\n| Pass results to an LLM inside your app | All eight APIs above | \n| Keep a user's results in their chat history or a shared transcript | Perplexity, Tavily, SerpApi, Serper, cloro. Brave only transiently. Parallel only for that user. Exa needs written permission | \n| Cache results and serve them to other users | No clause against it at Tavily, Perplexity, SerpApi, Serper or cloro. Parallel and Brave forbid it | \n| Fine-tune or evaluate a model on results | Not Brave, not Parallel, not Gemini grounding | \n| Build a long-term dataset | Brave custom plan, or a written agreement. Read the source's terms too | \n\nIf the agent only needs a cited answer for one user, any of these works, the built-in tools included. If you plan to keep results, check this table first, and then ask the vendor in writing. Simon Willison asked the same thing [on Hacker News](https://news.ycombinator.com/item?id=49965369): can you \"store and resyndicate results\"? The terms he quoted there, from another search API, allowed storage only as \"reasonably necessary to display such Output\", which he called \"infuriatingly vague\". A yes in an email is more certain than a clause like that.\n\nThis comparison is published by cloro, which is one of the eight APIs in the first table. Its row comes from its own published terms and is read the same way as the others. Each vendor's terms page was opened on November 16, 2026, and every clause is quoted from that page. \"No clause found\" means the current terms have no clause on that point; it does not mean the use is allowed by law or by the source website. Exa's terms are a PDF with no date. Parallel's Customer Terms have no date and apply in \"the version ... in effect at the time you became a customer\". Enterprise contracts can change any row. The evidence for the question came from Hacker News threads on [display rules](https://news.ycombinator.com/item?id=43920781) and the [SerpApi lawsuit](https://news.ycombinator.com/item?id=46329109).", "url": "https://wpnews.pro/news/which-search-api-lets-me-store-results-and-use-them-with-an-llm", "canonical_source": "https://dev.to/cloro/which-search-api-lets-me-store-results-and-use-them-with-an-llm-1g32", "published_at": "2026-11-16 11:13:32+00:00", "updated_at": "2026-11-16 11:23:04.719409+00:00", "lang": "en", "topics": ["ai-tools", "ai-search", "ai-policy"], "entities": ["Perplexity", "Tavily", "SerpApi", "Serper", "Brave", "Parallel", "Exa", "OpenAI"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/which-search-api-lets-me-store-results-and-use-them-with-an-llm", "markdown": "https://wpnews.pro/news/which-search-api-lets-me-store-results-and-use-them-with-an-llm.md", "text": "https://wpnews.pro/news/which-search-api-lets-me-store-results-and-use-them-with-an-llm.txt", "jsonld": "https://wpnews.pro/news/which-search-api-lets-me-store-results-and-use-them-with-an-llm.jsonld"}}