Settings - Text2SQL

• Always Invoke Text2SQL

ON/OFF Toggle

OFF

If enabled, the “text2SQL” translation will be performed for every question posed in Easy Answers. If disabled, the system will reuse previous natural language questions and their translations into SQL (indexed by Elasticsearch) from the cache. In a production environment, this toggle should be disabled to gain performance. Always Invoke Text2SQL is disabled by default.

• Enable Auto Spell Correction

ON/OFF Toggle

ON

If enabled, when entering natural language questions (NLQs) in Easy Answers, an automatic spell-check correction will be performed. This setting can be overridden by individual users by using a similar setting in the User Settings dialog available from the User menu. If disabled, the Auto Correction will not be performed, and the User Settings override toggle will not be available.

• Enable Narration Streaming

ON/OFF Toggle

ON

If enabled, the processing done (response tokens generated) by the LLM when interpreting questions will be streamed in real time into the Easy Answers UI.

• Stream LLM Output

ON/OFF Toggle

OFF

When on, also streams the raw model output live under the narration lines. Raw output can quote prompt fragments and schema internals, so it stays off.

• Default LLM Mode

Dropdown

Balanced

Select from the dropdown which LLM Interpretation Mode users in Easy Answers will use when asking questions, including: Controlled, Balanced, and Freeform. Users with permissions to change the LLM Interpretation Mode at runtime will be able to select any of the three modes when retrying a question.

• Number of LLM Execution Threads

Number Field

1

With the LLM Validation Strategy set to Default, only 1 thread can be configured. If the LLM Validation Strategy is Strict, then 2 or more LLM Execution Threads can be configured.

• Cache LLM Raw Prompt Responses

ON/OFF Toggle

ON

If enabled, it will cache raw LLM prompt responses to improve performance.

Question Rephrasing Model

Dropdown

Use Default

Select the LLM Model to use for Question Rephrasing.

• See Large Language Models Available in the AO Platform.

Question Intent Model

Dropdown

OpenAI ChatGPT 6 Luna No Reasoning

Select the LLM Model that will be used to rewrite a question into its abstract, value-agnostic intent - store it on each semantic template and generate it for every incoming question to retrieve matching templates by meaning, relative to the ontology being used.

• See Large Language Models Available in the AO Platform.

Ambiguity Model

Dropdown

OpenAI ChatGPT 6 Sol no Reasoning

Select the LLM Model to use for the Ambiguity Model.

• See Large Language Models Available in the AO Platform.

Ambiguity Clarification Resolution Model

Dropdown

OpenAI ChatGPT 6 Luna No Reasoning

Select the LLM Model to use for the Ambiguity Clarification Resolution Model.

• See Large Language Models Available in the AO Platform.

Enable LLM NER Extraction

ON/OFF Toggle

ON

If enabled, named entities are extracted with the Global NER prompt instead of the existing NLP extraction flow.

• NER Extraction Model

Dropdown

OpenAI ChatGPT 6 Luna No Reasoning

Select the LLM Model to use for the NER Extraction Model.

• See Large Language Models Available in the AO Platform.

Enable Prompt Token Tagging

ON/OFF Toggle

OFF (fixed)

If enabled, the Prompt token tagging strategy is included in token matching, letting the LLM tag tokens that the other strategies did not resolve.

• Question Tagging Model

Dropdown

Use Default

Select the LLM Model to use for Question Tagging Model.

• See Large Language Models Available in the AO Platform.

Enable Unrecognized Token Filter

ON/OFF Toggle

ON

If enabled, the unrecognized token filter matcher runs as the last token matching strategy, using the LLM to mark leftover tokens as stopwords, operators, literal values, or merged phrases. If disabled, the call is skipped, and the unrecognized tokens are left as the other strategies produced them. Enabled by default.

• Unrecognized Token Filter Model

Dropdown

Use Default

Select the LLM Model to use for the Unrecognized Token Filter Model. This configuration will be used to filter unrecognized tokens and identify operators in questions relative to the ontology being used.

• See Large Language Models Available in the AO Platform.

Enable Semantic Templates

ON/OFF Toggle

ON

If enabled, the Semantic Template settings will be available for configuration and used when creating and matching templates with questions in Easy Answers, or when managing templates on the Semantic Templates page in the Ontology Composer.

When disabled, template matching and template-based SQL generation are bypassed, and Generate Template on Verification is forced off. All other configuration properties will not be shown.

Enable Early Template Match

ON/OFF Toggle

ON

If enabled, a “fast path” is selected; before token identification, reuse the closest approved template when the question differs only by values/named entities (gated by question similarity and an intent check).

• Enable Early Exact Match

ON/OFF Toggle

ON (follows Enable Early Template Match)

Takes the value of Enable Early Template Match. Deterministic no-LLM fast path: reuse an approved template whose entity signature exactly matches - verbatim or with a deterministic limit swap. Anything needing an LLM falls through. Runs before Early Template Match.

• Enable Coverage Exact Match

ON/OFF Toggle

ON (follows Enable Early Template Match)

Takes the value of Enable Early Template Match. Widens Early Exact Match to every question an approved template has already covered. Only applies when Early Exact Match is on.

• Max Early Match Candidates

Number Field

10

Retrieved candidates the catch-early fast path runs its gates (Jaccard, value-delta, intent) against, highest-ranked first.

• Max Changed Words Allowed

Number Field

2

Fallback for the catch-early similarity gate: how many words may differ when the Jaccard score falls short. Counted after filter values are masked out.

Enable Hybrid Search Matching

ON/OFF Toggle

ON

If enabled, templates are matched using hybrid (lexical + semantic) search.

• Semantic Similarity Threshold for Hybrid Match Strategy

Number Field

0.6

Minimum semantic similarity score (0-1) required for a question to match under the hybrid match strategy. Higher values yield stricter, more accurate matches.

• Max Number of Similar Questions for Hybrid Match Strategy

Number Field

10

Maximum number of similar questions returned when using the hybrid match strategy.

Enable Semantic Signature Matching

ON/OFF Toggle

ON

If enabled, templates are matched using semantic signatures.

• Enable Topic Signature Matching

ON/OFF Toggle

OFF (hidden)

If enabled, templates are matched using topic signatures.

• Parallel Intent Validation

ON/OFF Toggle

ON (fixed)

Validates the shortlisted templates concurrently for lowest latency (spends the whole shortlist’s LLM calls). Off validates sequentially, stopping at the first confirmation. The matched template is identical either way.

• Max Intent Match Candidates

Number Field

3

Retrieved candidates that the hybrid strategy runs LLM intent validation against, highest-ranked first.

Execution Mode

Dropdown

Auto

Auto picks between Cost and Latency from the ontology’s measured template hit rate: a low hit rate favors speculating, a high one favors waiting for the match. Cost runs template match first and generates only on a miss. Latency starts generation speculatively alongside the match. The answer is identical; only timing and spend differ.

• Speculative Generation Hedge Delay (ms)

Number Field

1500

Latency/Auto modes only: how long the template match is given to settle before speculative generation starts. Ignored in Cost mode.

• Auto Mode Max Cost Increase (%)

Number Field

10

Auto mode only: the most extra LLM spend Auto may accept to buy the latency saving.

• Auto Mode Min Latency Saving (%)

Number Field

10

Auto mode only: the smallest latency improvement worth paying for.

• Return Template Edits Instead Of Full SQL

ON/OFF Toggle

ON (fixed)

On whenever Enable Semantic Templates is on. On a template reuse, the model returns the change as literal find/replace edits rather than rewriting the whole query; edits are applied in code with fallback to full generation. Takes effect only where the edit prompt is seeded for the configured model (GPT 5.6).

• Best Template Selector Model

Dropdown

GPT 6 Luna No Reasoning

Select the LLM Model to use for the Best Template Selector Model. Lightweight model that picks the fallback template best matching a question intent from the hybrid-search candidates.

• See Large Language Models Available in the AO Platform.

• Template Validator and Query Generator Model

Dropdown

GPT 6 Sol No Reasoning

Select the LLM Model to use for the Template Validator and Query Generator Model. Validates a matched template and regenerates its Semantic SQL for a new question. This configuration will be used to validate a matched template and regenerate its Semantic SQL for a new question relative to the ontology being used.

• See Large Language Models Available in the AO Platform.

Use Search Store for Token Matching

ON/OFF Toggle

OFF

If enabled, the Vector Store will be used for Token Matching. If disabled, the internal AO platform logic will be utilized.

• Valid POS Tags for Search Store for Token Matching

Dropdowns

Noun, Adjective

Repeater section for POS Tags. If enabled, use Search Store for Token Matching. If disabled, use internal AO Platform logic. Dropdown includes NOUN, ADJECTIVE, VERB, and ADVERB.

Enable Token Reuse from Templates

ON/OFF Toggle

ON (fixed)

If enabled, token matching first reuses the MSO and MSO-property associations of an already-approved semantic template whose intent matches the question, and only the spans it cannot resolve go on to graph, entity, and LLM token matching. Values, entities, and rules are never reused. Requires Enable Semantic Templates and Enable Hybrid Search Matching, which supply the intent-based retrieval it uses. Enabled by default.

Token Reuse Agreement Candidates

Number Field

3

How many of the intent-refined candidate templates must agree before an association is reused. The best candidate supplies the mapping; any other candidate within this many that maps the same wording to a different MSO or property vetoes it, so a span the candidates disagree about is resolved by token matching instead of by whichever template ranked first.

Run Token Reuse Before Graph Matching

ON/OFF Toggle

OFF (fixed)

Where token reuse sits relative to graph token matching. Off (default) runs graph first, because it resolves the same MSO and MSO-property associations deterministically and more cheaply, and a question it answers in full never reaches reuse at all - what reuse saves is the entity and LLM matching that follows either way. On runs reuse first, which only pays off where it resolves spans the graph could not. Both orders are correct; this exists so the two can be compared.

Semantic Similarity Threshold for Token Reuse

Number Field

0.75

How similar a stored question must be to the asked one before its token associations may be reused, as a raw question-to-question vector similarity between 0 and 1. Deliberately separate from the hybrid match threshold, which is measured between generated intents and is therefore a different score distribution. Higher by default, because a reused association has nothing downstream validating it, whereas a matched template still faces intent validation.

Enable Hybrid Search Token Context

ON/OFF Toggle

ON

If enabled, hybrid search narrows the ontology context sent to the token-matching LLM to the most relevant MSOs and properties; if disabled, the full ontology context is sent.

• Minimum Hybrid Search Score

Number Field

0.3

Minimum hybrid-search relevance score (0-1) for an MSO or property to be added to the token-matching context. Deterministically identified MSOs/properties are always kept regardless of this threshold.

• Maximum MSO Candidates

Number Field

7

Maximum number of hybrid-search MSO candidates added to the token-matching context (the deterministic floor is not counted against this cap).

• Maximum Property Candidates

Number Field

15

Maximum number of hybrid-search property candidates added to the token-matching context.

Use Vector Store for Token Matching

ON/OFF Toggle

OFF

If enabled, the Vector Store is used for token matching. This section only shows if the “Use Vector Store” configuration is enabled in the Vector Store section.

• Similarity Threshold for Token Matching

Number Field

Unset - platform default

The Similarity Threshold for Token Matching value determines the results returned by the Vector Store in response to the “Did you mean…?” functionality in Easy Answers. “Did you mean…?” is a way of resolving partially understood questions based on previously asked questions. The higher the threshold, the more accurate (but likely fewer) alternative question suggestions will be shown.

• Similarity Threshold for Nouns in Token Matching

Number Field

0.7

The Similarity Threshold for Nouns value determines the results returned by Vector Store in response to the “Did you mean…?” functionality in Easy Answers. “Did you mean…?” is a way of resolving partially understood questions based on previously asked questions. The higher the threshold, the more accurate (but likely fewer) alternative question suggestions will be shown.

• Match for Token Types

Text Field

mso, mso_property

• mso

• mso_property

• Bypass LLM Token Matching

ON/OFF Toggle

ON

If enabled, the LLM token-matching step is skipped, and the hybrid-search candidates above are sent to Text2SQL as schema context instead. Saves two LLM calls on questions the other strategies could not fully resolve, but those spans are no longer mapped to a specific table, column, or value - so filters that depend on a matched value may be lost, and Text2SQL sees a broader, less precise context. Leave off unless token matching is too slow.

Custom Entities to be Included

Dropdowns

(none)

Repeater section - The Custom Entities to be Included refers to configuring how custom entities are recognized.

Use Vector Store

ON/OFF Toggle

OFF

If enabled, this will enable the selected Vector Store to identify “Did you mean” alternative question suggestions in Easy Answers. If disabled, the Vector Store is not used for Text2SQL.
Note: All the following configuration properties will only become available if this property is enabled.

• Vector Store

Dropdown

Use Default

Use the dropdown to select from the available Vector Store:

• Use Default

• Pinecone DB

• Weaviate DB

• Embedding Model

Dropdown

All Mpnet Base V2

Embedding involves mapping high-dimensional vectors into a lower-dimensional space, facilitating machine learning tasks with large inputs, such as sparse vectors that represent words. Use the dropdown to select from available Embedding Model configurations.

• See Embedding Models Available in the AO Platform.

• Vector Matching Algorithm

Dropdown

Cosine Similarity

Algorithm used to match vectors. The dropdown includes: Dot Product, L1 Distance, L2 Distance, Euclidean, Cosine Similarity.

Enable Easy Answers Suggestions

ON/OFF Toggle

ON

If enabled, the “Did you mean…?” functionality will be enabled in Easy Answers. If disabled, then the Easy Answers alternative question suggestions will not appear for user questions that are not fully understood.

• Single Token Similarity Threshold

Number Field

0.82

The Single Token Similarity Threshold value is used to suggest individual Words for the “Fix Unknown Words” functionality in Easy Answers. In that workflow, individual words will be suggested for words not understood by our Natural Language interpretation. The higher the threshold, the more accurate (but likely fewer) word suggestions will be shown.

• Similarity Threshold

Number Field

0.80

The Similarity Threshold value determines the results returned by Vector Store in response to the “Did you mean…?” functionality in Easy Answers. “Did you mean…?” is a way of resolving partially understood questions based on previously asked questions. The higher the threshold, the more accurate (but likely fewer) alternative question suggestions will be shown.

• Suggestion Count

Number Field

3

The Suggestion Count determines the maximum number of suggestions shown in Easy Answers.

Enable Knowledge Base Indexing

ON/OFF Toggle

OFF

This is the master switch for Knowledge Base ingestion of templatized questions. Every other setting in this section is honored only while this is on.

• Ingestion Model

Dropdown

Use Default

Extracts the reusable knowledge (topics and tokens) from a question and its answer.

• Token Review Model

Dropdown

Use Default

Reviews the extracted tokens before they are persisted.

• Topic Refresh Model

Dropdown

Use Default

Refreshes topic descriptions after new tokens are assigned to them.

• Topic Assignment Model

Dropdown

Use Default

Assigns each extracted token to a candidate topic.

• Pseudo SQL Model

Dropdown

Use Default

Derives the reusable pseudo-SQL for a token.

• Token Extraction Model

Dropdown

Use Default

Token and context breakdown - one call per candidate, so a smaller model is usually the right choice.

• Finalizer Model

Dropdown

Use Default

Retained for contract compatibility; the finalizer phase is no longer invoked.

• Duplicate Token Model

Dropdown

Use Default

Duplicate-token detection model; consulted only when Duplicate Token Check is on.

• Enable Duplicate Token Check

ON/OFF Toggle

OFF

If enabled, checks newly extracted tokens against existing ones for duplicates before indexing.

• Embedding Model

Dropdown

(none)

Leave blank to use the model the knowledge indices are embedded with; query and document vectors must come from the same model.

• Self-Consistency Threads

Number Field

3

Self-consistency vote count during knowledge ingestion.

MSO Identification Model

Dropdown

Use Default

Select the LLM Model to use for MSO Identification.

• See Large Language Models Available in the AO Platform.

MSO Property Identification Model

Dropdown

Use Default

Select the LLM Model to use for MSO Property Identification.

• See Large Language Models Available in the AO Platform.

Enable Schema Pruner

ON/OFF Toggle

ON (fixed)

Reduces the schema context to the tables, columns, and relationships the generated Semantic SQL actually uses. Runs in the background and never delays the answer - the full schema context is used when a pruned one is unavailable.

Schema Pruner Model

Dropdown

OpenAI ChatGPT 5.6 Luna Low Reasoning

Select the LLM Model to use for the schema-pruning pass.

• See Large Language Models Available in the AO Platform.

Schema Pruner Minimum Context Size

Number Field

2000

Smallest schema context (characters) worth pruning; below this, the pruning call costs more than it saves.

Enable Prompt Atom Routing

ON/OFF Toggle

OFF (fixed)

If disabled, every curated ontology, entity, and property instruction for the entities a question touches is sent with that question. Enable to send only the instruction blocks the question actually needs, selected by search. Requires the ontology prompts to have been atomized first.

• Routed Instructions Processor Model

Dropdown

Use Default

Select the LLM Model to use for Routed Instructions Processor Model. Applies routed instructions to a generated statement when the pipeline step that would normally carry them is off.

• See Large Language Models Available in the AO Platform.

• Prompt Atomization Model

Dropdown

Use Default

Select the LLM Model to use for Prompt Atomization Model. Breaks a curated instruction into atomic prompt blocks. Runs on prompt save / re-index, never while answering.

• See Large Language Models Available in the AO Platform.

• Prompt Block Split Model

Dropdown

Use Default

Select the LLM Model to use for Prompt Block Split Model. Proposes cut points for oversized instruction blocks that could not be split on structure alone.

• See Large Language Models Available in the AO Platform.

Include MSO Schema in Prompt

ON/OFF Toggle

OFF

If enabled, the MSO schema is included in the prompt.

Use Statistics for LLM Content Generation

ON/OFF Toggle

OFF

If enabled, the Statistics gathered during MSO Discovery will be used to generate LLM Content.

Use Statistics for Semantic SQL Optimization

ON/OFF Toggle

OFF

If enabled, the Statistics gathered during MSO Discovery will be used when performing semantic queries, ensuring they scale efficiently with the data.

Use LLM Content from Parent MSO

Dropdown

Ignore Parent Content

Select from the following inheritance options:

• Use Parent Content - will copy LLM Content from the referenced (parent) ontologies. The user can edit the content, but if LLM Content is changed in the referenced ontology, it will override the current ontology.

• Ignore Parent Content - no LLM Content will be referenced or copied from the referenced (parent) ontologies. The user shall generate LLM Content for the child ontology.

• Copy Parent Content - will copy LLM Content from the referenced (parent) ontologies. The user can edit the content independently of the LLM Content in the referenced ontologies.

MSO Prompt Priority

Dropdown

Low

Dropdown includes: Low, None, Never.

The MSO Prompt Priority can be auto-populated using the Generate LLM Content from the Options menu > Advanced in both Ontology and MSO Composers. Viewing or manually creating MSO Prompt Priorities can be done for each MSO on the LLM Content page. See MSO - LLM Content.

MSO Property Prompt Priority

Dropdown

Low

Dropdown includes: Low, None, Never.

The MSO Property Prompt Priority can be auto-populated using the Generate LLM Content from the Options menu > Advanced in both Ontology and MSO Composers. Viewing or manually creating MSO Property Prompt Priorities can be done for each MSO on the LLM Content Page. See MSO - LLM Content.

Enable Combined Reasoning and Query Generation

ON/OFF Toggle

ON

If enabled, the Semantic SQL reasoning and query generation are performed in a single combined LLM call.

• Combined Reasoning and Query Model

Dropdown

OpenAI ChatGPT 5.6 Terra Low Reasoning

Select the LLM Model used for the combined reasoning + generation call.

• See Large Language Models Available in the AO Platform.

• Merge Display Columns Into Generation

ON/OFF Toggle

ON

If enabled, the generation prompt projects the display/proof columns itself; the separate Additional Columns step runs only as a fallback when its pre-check finds columns still missing.

• Semantic SQL Reasoning Generator Model

Dropdown

Use Default

Select the LLM Model to use for Semantic SQL Reasoning Generator.

• See Large Language Models Available in the AO Platform.

• Semantic SQL Generator Model

Dropdown

Use Default

Select the LLM Model to use for Semantic SQL Generator.

• See Large Language Models Available in the AO Platform.

Heuristic Enhanced Criteria

ON/OFF Toggle

ON

If enabled, the operator implied by the question (contains, starts with, equals, negations) is decided by a deterministic phrase table instead of an LLM call. Ambiguous phrasings keep the configured match type.

Apply Default Criteria

Dropdown

As part of Post-Processing as an LLM call

Default criteria to be used for LLM Strategy.

Use Heuristic for Additional Columns

ON/OFF Toggle

ON

If enabled, runs first and completes the key and name properties deterministically with no LLM call. The LLM pre-check and additional-columns call (the two models below) then run only for what is still missing.

• Additional SQL Columns Model

Dropdown

OpenAI ChatGPT 5.6 Luna No Reasoning

Select the LLM Model to use for the Additional SQL Columns Model. This configuration will be used to add safe display and proof columns to the generated Semantic SQL relative to the ontology being used.

• See Large Language Models Available in the AO Platform.

Additional Columns As Edits

ON/OFF Toggle

ON

If enabled, the additional-columns LLM call returns find/replace edits instead of rewriting the query. Edits must be additive and anchor unambiguously; otherwise, the call falls back to a full rewrite.

• Check Additional Columns Required Model

Dropdown

OpenAI ChatGPT 5.4 Nano Low Reasoning

Select the LLM Model to use for the Check Additional Columns Required Model. This configuration will be used to decide whether additional display columns need to be added to the generated Semantic SQL relative to the ontology being used.

• See Large Language Models Available in the AO Platform.

Enable Post Processing

ON/OFF Toggle

ON

If enabled, the generated Semantic SQL is put through an additional LLM post-processing step. Disable to use the query as generated and save that LLM call.

• Semantic SQL Post Processing Model

Dropdown

Use Default

Select the LLM Model to use for the Semantic SQL Post Processing Model.

• See Large Language Models Available in the AO Platform.

Enable Query Correction

ON/OFF Toggle

ON

If enabled, additional LLM requests are made to ensure optimal Text2SQL conversion.

• Semantic SQL Corrector Model

Dropdown

OpenAI ChatGPT 6 Luna Low Reasoning

Select the LLM Model to use for the Semantic SQL Corrector Model.

• See Large Language Models Available in the AO Platform.

• Query Corrector Repair Model

Dropdown

OpenAI ChatGPT 6 Luna Low Reasoning

Select the LLM Model to use for

• See Large Language Models Available in the AO Platform.

Parallel SQL Correction

ON/OFF Toggle

ON (fixed)

Runs the optional SQL passes (query corrector, statistics optimizer, default criteria, post-processing, additional columns, AO functions) concurrently over one SQL snapshot and merges their edits with the Merge SQL Edits Model. Faster, but uses all the passes’ LLM calls. Note: nothing runs when Enable Query Correction is off.

• Semantic SQL Functions Model

Dropdown

Use Default

Select the LLM Model to use for Semantic SQL Functions Model.

• See Large Language Models Available in the AO Platform.

Enable Query Functions

ON/OFF Toggle

ON

If enabled, MSO query functions are applied to the generated Semantic SQL through an additional LLM call. Disable to skip that step.

• AO Functions Gate Model

Dropdown

OpenAI ChatGPT 5.6 Luna Low Reasoning

Select the LLM Model to use for the AO Functions Gate Model.

• See Large Language Models Available in the AO Platform.

• Apply AO Functions Model

Dropdown

OpenAI ChatGPT 5.6 Luna Low Reasoning

Select the LLM Model to use for the Apply AO Functions Model.

• See Large Language Models Available in the AO Platform.

• Merge SQL Edits Model

Dropdown

OpenAI ChatGPT 6 Luna No Reasoning

Select the LLM Model to use for the Merge SQL Edits Model.

• See Large Language Models Available in the AO Platform.

• Functions Applicable for Ontology

Dropdown

aoWithinDuration, timestampAdd, aoFormatDate, timestampDiff, extract, date_trunc, percentile_cont, correlation

Repeater section. The LLM functions that may be applied for this ontology. Add one row per function; only the functions listed here are considered during Semantic SQL generation.

Semantic SQL Validator Model

Dropdown

Use Default

Select the LLM Model to use for Semantic SQL Validator.

• See Large Language Models Available in the AO Platform.

Query Verification Model

Dropdown

OpenAI ChatGPT 5.6 Luna No Reasoning

Verifies the generated Semantic SQL against the question when it contains unrecognized phrases and produces the message shown to the user. Runs on the answer path, so a small, fast model is appropriate.

LLM Validation Strategy

Dropdown

Default

Dropdown includes:

• Default - only validates that SQL syntax is executable.

• Strict - will include a specific LLM-based Query Validator (validating syntax correctness, areas are proper, conditions are met, performance efficiency) for n number of different variations of a question, seeking to optimize for the best possible answer.

• Validation Cutoff Score

Number field

90

Cutoff score for the LLM validation strategy is configured to Strict.

Validate Relationships

ON/OFF Toggle

OFF

If enabled, Semantic SQL relationship validation is enabled.

Test Query Through Execution

ON/OFF Toggle

OFF

If enabled, Queries are test-executed before delivery.

Enable Semantic SQL Metadata Extraction

ON/OFF Toggle

ON

If enabled, Semantic SQL metadata is extracted heuristically; the extraction LLM call is made only when this is off.

Semantic SQL Metadata Model

Dropdown

Use Default

Dedicated model for generating Semantic SQL metadata. Moved here from the Easy Answers Agent tab and now saved with the Text2SQL settings; a value saved earlier on the agent tab is still honored until this one is set.

Semantic SQL Sourcing Extractor Model

Dropdown

Use Default

Select the LLM Model to use for Semantic SQL Sourcing Extractor.

• See Large Language Models Available in the AO Platform.

Show Semantic SQL Reasoning and Sourcing

ON/OFF Toggle

ON

If enabled, it will make the Semantic SQL Reasoning and Sourcing functionality available in the Easy Answers Options menu on the Results page. See Viewing Sourcing Information | AO Easy Answers.

Show Data Summary

ON/OFF Toggle

ON

If enabled, a Summary of the question will be shown in Easy Answers.

Data Summary Generator Model

Dropdown

OpenAI ChatGPT 6 Luna No Reasoning

Select the LLM Model to use for generating Data Summary.

• See Large Language Models Available in the AO Platform.

• Enable Auto Height for Summary

ON/OFF Toggle

ON

If enabled, the Summary panel height adjusts automatically.

• Max Records for Data Summary

Number Field

100

Maximum records considered while generating the data summary.

Show Summary Insights Infographic

ON/OFF Toggle

OFF

If enabled, Summary Insights Infographics of the question will be shown in Easy Answers.

• Summary Insights Infographics Model

Dropdown

OpenAI ChatGPT 5.4 Mini No Reasoning

Select the LLM Model to use for generating the Insights Infographics Summary.

• See Large Language Models Available in the AO Platform.

Cache Data Summaries

ON/OFF Toggle

ON

If enabled, the Data Summary for the Semantic SQL will be cached.

• Response Summarization Model

Dropdown

Use Default

Select the LLM Model to use for generating Response Summarization.

• See Large Language Models Available in the AO Platform.

Show Next Question Suggestions in Easy Answers

ON/OFF Toggle

ON

If enabled, the Next Question Suggestions will be generated in the Chatbot based on the History list of past questions.

Show Filter Criteria for Easy Answers

ON/OFF Toggle

OFF

If enabled, Filter Criteria are shown in Easy Answers.

Show Apps for Generative AI Query

ON/OFF Toggle

ON

If enabled, additional Apps, if any, are shown along with the AI Response table.

Show Apps for Non-Passthrough Federated Query

ON/OFF Toggle

OFF

If enabled, Apps are shown for non-passthrough federated queries.

Generate Template on Verification

ON/OFF Toggle

ON

If enabled, a template is generated as soon as a question is verified. When disabled, templates are generated by a scheduled job (see Templatization Job Interval).

Run Templatization Job

ON/OFF Toggle

OFF

If enabled, a scheduled job templatizes the eligible questions of this ontology at the Templatization Job Interval. No job is scheduled until this is turned ON. Only applies when Semantic Templates is enabled.

• Templatization Job Interval

Text Field

0 0 * * * ?

Cron expression for the scheduled templatization job. Applicable only when Generate Template on Verification is disabled. Reference: https://crontab.guru/#*/3_*_*_*_*.

Auto Approve Templates

ON/OFF Toggle

ON

If enabled, generated templates are approved automatically instead of waiting for manual review.

Promote Variant Templates

ON/OFF Toggle

ON

If enabled, when a question answered via the slower hybrid path only adds value-bearing filters/properties to the template it reused, an independent, implicitly verified template is created asynchronously so that future variations take the fast exact-match path.

• Allow Promotion for Status

Dropdown

Verified Only

Which served variants may be promoted to their own template.

• Verified Only - only variants of a verified template.

• Allow Successful Questions - the base template need not be verified, but the question must have a run without error (status Success).

• Allow All - any variant, regardless of verification or execution status.

Templatize Error Questions

ON/OFF Toggle

OFF

If enabled, a question’s execution status is ignored while templatizing, so Failed and Partial questions can become templates too. All other exclusions (already templatized, no tokens, no semantic SQL, no category) still apply. Disabled by default; it templatizes only Success questions.