Linkup review

Linkup is a hosted web search API for AI applications. Its API covers web search, URL extraction and asynchronous deep research, returning sourced answers and full-text snippets for use in agents, RAG systems and other applications that need current web context. It sits between your application and web data rather than replacing your application’s search or orchestration layer.

The vendor also offers private indexes of proprietary data and a bring-your-own-cloud option. Per the homepage, that option runs the indexing layer inside your infrastructure; it is not a statement that the entire Linkup service can be self-hosted. Teams can tune the index, sources and freshness. For enterprise customers, Linkup lists personalized indexes and a private environment. This makes it relevant to teams that need to combine public web results with controlled data, while leaving the application and its agent logic in their existing stack.

The integration surface is an API, with separate Fetch, Search and Research capabilities. The homepage links to API docs and shows a Python example using LinkupClient; it also names Python, Node.js and cURL. The example sets an API key directly or through the LINKUP_API_KEY environment variable, then calls client.search with a query, depth and output type.

For orchestration, Linkup says it works with the OpenAI SDK wrapper, LangChain, CrewAI and other frameworks, and provides an integrations page. The homepage shows API calls and framework integrations rather than a CLI workflow, so teams that require a command-line tool should confirm current support in the vendor’s docs. The docs should also be the source of truth for request and response schemas, authentication details and error handling. Search responses can provide cited results and snippets; research is asynchronous, so applications need to handle completion rather than assume every call returns immediately.

Linkup charges per request, with no monthly or annual subscription billing listed. The pricing page marks a free plan, but does not publish its included quota or where the free allowance stops. Startups can apply for $5,000 in free credits; this is not an automatic credit grant.

Fetch Standard / Pro costs $0.001–$0.006 per request, with synchronous responses and latency under two seconds. Search Fast / Standard / Deep costs $0.005–$0.006 per request, with synchronous responses and stated latency of one to three seconds. Research costs $0.25–$2.50 per request and runs asynchronously, with stated latency of one to ten minutes. At those rates, 1,000 Search requests cost $5–$6 before any credits or enterprise terms.

Enterprise pricing is custom and includes personalized indexes, a dedicated index refresh rate, a private environment, ZDR, SOC 2 Type II and an SLA. The pricing page does not list request quotas or overage terms, so ask about rate limits and expected volume before budgeting production traffic.

Linkup presents a managed API and an option to run its indexing layer inside your own cloud. That is a deployment choice, not evidence that the full service is self-hostable. The vendor doesn't publish its software licence or source availability, so teams requiring open-source components should confirm those terms directly.

The private-index and bring-your-own-cloud options give enterprises more control over proprietary data, but the vendor does not describe a data-export path. Ask what index data and query history can be exported, in which formats, and whether the BYOC deployment changes those terms. If you migrate away, plan to replace Linkup API calls and adapt response handling in the parts of your application that consume its search results, citations, snippets or research output. Keeping that integration behind an internal interface can limit the cost of switching providers.

The homepage claims 99.9% uptime with an SLA and describes sub-second latency; the pricing page gives different latency ranges by product: under two seconds for Fetch, one to three seconds for Search, and one to ten minutes for asynchronous Research. Treat those as vendor-stated figures, not guarantees for every workload. Research’s asynchronous behavior also means your application needs a path for pending work and delayed results.

Before launch, confirm rate limits, timeout and retry guidance, index refresh behavior, and what happens when a request fails. Enterprise terms list ZDR, SOC 2 Type II and an SLA; ask what applies to your plan and deployment. The homepage describes BYOC as keeping the indexing layer inside your infrastructure, but teams should verify the data flow for each feature they use.

Reddit feedback is limited and mixed: one thread says Linkup has connections to premium sources, while another commenter preferred a competitor’s retrieval. These are individual opinions, not a broad reliability assessment.