Oracle Trusted Answer Search
Turn natural-language questions into trusted application actions.
Oracle Trusted Answer Search enables teams to build secure, accurate, and fast natural-language interfaces, matching domain-specific questions to application-specific targets, ensuring quick outcomes without risking data exposure or unpredictable responses.
Make existing applications easier to use
Give users a natural-language path to the capabilities your applications already provide, while preserving governance, authorization, and control over every result.
Increase user self-service
Help users discover existing reports, pages, and workflows using familiar business language and cut their "where do I find?" requests
Reduce support effort
Give users consistent, reliable answers from your curated reports that they can trust. Easily fine-tune search results to improve relevance.
Keep outcomes governed
Return curated targets and validated inputs instead of free-form answers, while the application continues to enforce authorization.
How Trusted Answer Search works
1
Ask naturally
A user asks, “I lost my debit card while traveling yesterday. How do I lock it and get a replacement?”
2
Match and resolve
Hybrid lexical and vector retrieval finds the most relevant target - "Lock Customer Account" to the user question and resolves inputs such as categories, time period, and IDs - the "DEBIT_CARD" category in this case
3
Take the trusted action
Trusted Answer Search returns the action, a URL or SQL query, corresponding to the approved target with the resolved inputs: "www.demo-brokerage.com/access/account/replace/type=DEBIT_CARD"
4
Improve continuously
Experts can influence the search results by providing feedback and reviewing telemetry, run evaluations, and publish the improved search configuration.
Product in action
Find the right destination by asking naturally
Users search in natural language and receive ranked, curated targets with resolved inputs. Treat your users with a delightful, fully-functional, customizable search widget integrated with the Trusted Answer Search backend.
Manage, test, release smartly
Search administrators and search-space experts manage targets, descriptions, sample questions, value sets, access, and versions in a dedicated Oracle APEX application. Search telemetry, end-user feedback, and tests help teams evaluate the search results.
Refine search with expert feedback
Help users find the right reports by reviewing search results and providing expert feedback to improve results for similar questions.
Explore how to guide search results
Validate improvements with repeatable tests
Run evaluation sets of questions and their ground truth to measure search accuracy and response times. Review individual outcomes and catch regressions before publishing changes.
Turn search activity into actionable insights
Monitor search engagement, explore query trends, review user feedback and unsuccessful searches to identify opportunities to improve relevance.
Release Highlights
Match queries with regular expressions (Regex)
Define regular expressions to route questions matching a specific phrase or identifier format directly to a chosen search target
Clone search spaces
Clone an entire search space to reuse an existing setup as the starting point for another application or use case
Flag questions and guide search results
Build on admin feedback by flagging questions for follow-up, adding review notes, and tracking resolution.
Monitor search outcomes and performance
Use search analytics and telemetry to track search target selections and unsuccessful searches. Review latency summaries and detailed breakdowns to understand search performance.
Reduce embedding model overhead
Reduce memory use and repeated model-loading overhead by sharing ONNX embedding-model components across concurrent database sessions. Learn more
Release Highlights
Sign in with your organization’s account
Use single sign-on (SSO) authentication to access Trusted Answer Search through your configured identity provider
Extract dynamic inputs from natural language
Use LLM target inputs to extract values from questions and pass them to the selected target, URL, or other target action
Improve search result rankings
Use LLM-based re-ranking to reassess retrieved results and place the most relevant matches higher in the result list
Embed search in your application
Add the ready-to-use Trusted Answer Search JavaScript widget to an existing application, bringing natural language search directly into the user’s workflow
Connect to large language models
Configure supported LLM models to extract target inputs and refine search result rankings. Learn more
Validate and improve search continuously
Support continuous improvement with repeatable evaluation sets to measure search quality, and use telemetry to understand real-world search behavior and guide further improvements
Release Highlights
Find matches by words and meaning
Use hybrid search to combine keyword and vector retrieval, matching questions with relevant search targets through both wording and meaning
Manage search changes with versions
Use search space versioning to create and compare versions of your search setup, then publish the version you want users to access
Improve search through feedback
Use feedback-based improvement to refine search behavior based on people’s feedback on results
Recognize different terms for the same value
Define target value sets with standard values and synonyms so questions using different terms resolve to the inputs your targets and applications expect
Control access with database privileges
Use database user-based authorization to manage access to Trusted Answer Search through database user identities and assigned TAS privileges
Product announcement
Announcing Oracle Trusted Answer Search 26.3Explore regex-based query routing, search space cloning, enhanced question review workflows, and deeper search analytics, alongside more efficient embedding-model resource use.
Learn More
- Product launchOracle introduces Trusted Answer Search
- PerspectiveEnterprise AI search needs a new approach