Skip to content

Seek

traced to their source.

Seek answers questions from your own documents. Every answer is scored, and the passages it came from are highlighted. Try it on the SaaS today.

Talk to an engineerSee the guardrails

How an answer is built

How Seek builds an answer

Each step is a setting you control. An answer that misses your thresholds goes to a fallback reply or agent instead.

  1. Hybrid search over your content

    Keyword, vector or hybrid search over your own knowledge base. You choose the query type, and custom vector queries work where your search backend offers them.

  2. Pick the right sources

    A score range decides which documents qualify, a date penalty pushes older ones down, and a cap sets how many documents reach the model.

  3. Re-rank by meaning

    Passages are re-ordered by how well each one matches the answer. Answers stitched from many documents score lower, and you set the minimum coverage for the top source.

  4. Check each sentence

    Names and search terms in the answer are checked against the sources. Sentences with key words your knowledge base doesn't contain can be removed before the answer goes out.

  5. Confidence thresholds you set

    Below a warning level, the answer gets a caveat. Below the minimum, your own reply or a fallback agent takes over. A separate level decides when to show a link.

  6. Sources shown with the answer

    Provenance highlighting shows which parts of an answer came from your sources. Each source is named, and its link is shown when confidence is high enough.

Grounded and cited

Every answer comes with its scores

Every answer shows its sources, its scores and how long each check took.

console › seek
A Seek answer with its source phrases highlighted, the session settings, a sentiment gauge and the answer's scores and timings.
A live answer in the Seek tab: source phrases highlighted, the session panel beside it, the per-answer metrics beneath.

Low-confidence answers

When Seek can't answer well

Below the minimum confidence you set, Seek gives your own reply or passes the question to a fallback agent instead of guessing. The log keeps the score, the topic and the reason.

Route it onward

console › governance › seek logs
The Seek logs under Governance: dated questions and answers, each with its topic, category, score and response time, one flagged for low confidence.
Seek Logs under Governance: every low-confidence answer is logged with its score.

The cache

Reuse answers you've approved

Questions are matched to topics, and a topic's approved answer can be served without calling the model. Each topic shows its coverage and confidence over time, flags source changes, and can be edited by your experts.

Matched to a topic

console › curate
The Curate tab: a table of topics, each with a category, question and answer counts, and coverage and confidence curves.
The Curate tab: topics with their question and answer counts, coverage and confidence.

Common questions about Seek

What teams ask once they've seen Seek answer their first questions.

  1. How do we know an answer is actually grounded?

    Seek measures it. Each answer carries a semantic match score, your knowledge base's confidence and coverage, and the source passages. Answers under your thresholds are replaced or routed, and Governance tracks the scores over time.

  2. What happens when your content doesn't have the answer?

    Seek says so. A minimum-confidence setting swaps a weak answer for your own reply, and a fallback agent can notify a team or open an issue. Either way, it's logged.

  3. Does caching mean we serve stale answers?

    Cached answers last as long as you set. Answers are flagged when their source documents change, and Curate lets you filter for edited, flagged, new and outdated answers so your team can keep them current.

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.

Start nowRequest a demo