Agents
Find the answer, not just the document
The Search Agent indexes your knowledge by meaning, answers questions with citations drawn from across systems, and never returns anything the person asking is not allowed to see.
What it does
Search by meaning, answer with sources
Keyword search fails when people do not know the exact words a document used. The Search Agent indexes your knowledge base by meaning, so a question finds the relevant passage even when the wording differs. It returns a direct answer with the sources it drew from, cited and linked, so the reader can confirm the answer rather than trust it — and it searches across silos in one place instead of one system at a time.
Semantic index
Vectorises content so meaning, not exact wording, drives what is found.
Answers with citations
Returns a direct answer and links every claim to its source passage.
Across silos
Searches many systems at once, so people stop hunting tool by tool.
Inputs it accepts
What you feed it
Documents and wikis
Policies, runbooks, wikis and files across your content stores.
Structured records
Governed data products and records that hold answers to factual questions.
Tickets and threads
Support tickets, chat threads and email where institutional knowledge hides.
Decisions it makes on its own
What it answers without asking
For a question it can answer from indexed sources, the agent does so directly, choosing the passages that best support the answer and enforcing the asker's permissions on every result. It keeps the index current as content changes, ranks sources by relevance and recency, and cites what it used so the answer stands on evidence rather than assertion.
Enforces access
Filters every result by the asker's permissions, so search never leaks restricted content.
Keeps the index fresh
Re-indexes as content changes, so answers reflect the current state.
Ranks by relevance
Prefers the most relevant and recent sources, and cites the ones it used.
What escalates to a human
Where it holds back
No supporting source
If nothing in the index supports an answer, the agent says so rather than inventing one, and points to who might know.
Conflicting sources
When credible sources disagree, it surfaces the conflict and the passages rather than picking a side silently.
Systems it connects to
Where it reads
Content stores
Document management, wikis and shared drives.
Collaboration tools
Chat, ticketing and email where knowledge accumulates.
Data products
Governed products for factual, structured answers with lineage.
A worked example
What is our refund policy for enterprise contracts?
An agent asks a question that touches three sources: a policy document, a signed contract clause and a past support ruling. The Search Agent finds all three by meaning, checks that the person asking may see the contract, and returns a single answer citing each source with a link. Because the contract clause is more specific than the general policy, it leads with that and notes where the two differ — rather than returning ten documents to read.
Questions
Frequently asked
- Can it surface content people should not see?
- No. Access is enforced on every result against the asker's permissions, so search respects the same rules as the underlying systems.
- How current are the answers?
- The index updates as content changes, so answers reflect the current state rather than a stale snapshot. Where sources conflict, the agent surfaces the conflict.
- What if there is no good answer?
- It says so and points to where the answer might be found, rather than fabricating one. Every answer it does give is backed by cited sources.
Search your own knowledge base
Connect a corner of your content and we will show you semantic answers with citations, and how access is enforced on every result.