Semantic retrieval
Where the first answer to any topic is built: hybrid search, accuracy checks, confidence thresholds and citations.
See how an answer is builtSemantic cache
you already trust.
Most questions your users ask have been asked before in other words. Approve the answer once and serve it again without another model call.
How it works
How the cache answers a question
A new question is matched to a topic, then answered from the cache. Neither step calls the model.
Matched to a topic
Each question is grouped under a topic, a set of questions with the same meaning. NeuralSeek generates similar questions for each one, so a familiar question in new words finds its topic by exact or fuzzy match.
Served from the cache
When an expert edits an answer, it can be cached and served directly, with no new generation. Normal answers have their own cache, behind a threshold you set. Seek labels these answers Cached.
Curated by your experts
A cache your team can read and edit
Cached answers live in Curate, with indicators, filters and search, so your experts can review what the cache serves.
Edited by your expert
Edit any generated answer for style and content, and the edit is marked. Upload curated questions and answers in bulk from a spreadsheet, and improve them through Seek first if you like.
Flagged per topic
Each topic carries indicators: whether it has new answers, whether it contains personally identifiable information, and whether the data it was answered from has gone out of date.
Out-of-date answers detected
Each cached answer carries a hash of its source passage, checked against the knowledge base. When a document changes, the answer is flagged as out of date.
Filtered and searched
Search topics by keyword and filter for the ones edited, newly answered, flagged or out of date. Your experts go straight to the answers that need them.
Measured
Cache results in Governance
The Governance tab reports the cache next to the rest of retrieval, filtered by topic, category or date.
Cache efficiency shows how often answers came from the cache, from an edited answer, or were generated new. Coverage and confidence are charted per topic over a lookback period, so drift is easy to spot.
Cache efficiency
console › governance › semantic insights
Where it sits
Part of Seek and Governance
The cache works between retrieval and governance. It serves the answers your experts edited and the answers that clear your threshold.
Agent workflows also have their own local cache: a key-value store an agent reads, writes and searches by index and key, including phonetically. It is separate from the answer cache.
Governance
Where the cache is measured with everything else: efficiency, coverage, confidence, tokens and cost.
See what is recorded
Start building with NeuralSeek today.
Sign up online and start with a two-week trial, or talk to a NeuralSeek expert about running it in your own cloud.