Mojeek: Independent Search Index and AI Summary
Mojeek is a UK search engine with its own index. Who runs it, how its AI Search Summary cites results, and its NoML proposal for machine learning.
Mojeek is a search engine from the UK with its own web index. Its optional Summary feature uses a large language model (LLM) to summarize the search results, with citations that link to those results. A page can be a source for a summary only if it is in the Mojeek index.
Who runs Mojeek
The about page names Marc Smith as founder and developer. The office is at the Sussex Innovation Centre in Brighton. The company says that it does not track its users. It also sells its results through a Web Search API. For a longer look at privacy and search features, see this review of the engine.
The search index
Work on the current index started from scratch in 2011. In 2025, the index passed 9 billion pages. The engine fills the index with its own crawler, so the results do not come from another search engine.
How Search Summary works
Search Summary launched on April 3, 2024, as the company blog announced. The feature uses retrieval-augmented generation (RAG). RAG means that the model writes its answer from documents it retrieved for the query.
- Mojeek runs the query against its own index.
- It sends the top results and the query to an LLM. At launch, this was Mixtral, accessed through Lepton.
- The LLM writes a summary. The summary shows beside the results on desktop, or in a separate Summary tab.
Each citation in the summary links back to a normal Mojeek result. Users can turn summaries on or off. The summary uses the normal results, so there is no separate crawler for it. The documentation gives no way to opt out of summaries only.
Which bots to allow or block
MojeekBot is the only crawler that Mojeek documents. Its guide gives the robots.txt rules, the robots meta tag values and the checks for real requests. A page that is not in the index cannot appear in a summary.
The NoML proposal
Mojeek proposed NoML, a robots meta value for pages that can appear in search but must not be used for machine learning (noml.info):
<meta name="robots" content="noml">The NoML page does not list any crawler that supports the value today. It is a proposal, not a standard.