[Inference] Prompts: reasoning mode + ES|QL prompt by dgieselaar · Pull Request #226616 · elastic/kibana

dgieselaar · 2025-07-04T16:22:22Z

Adds a utility function that will execute a Prompt (system message, user message template and tool definitions), in a way that the LLM will reason about the task and its progress and execute tools autonomously until it feels confident enough to answer the question. As an example, a reference implementation was added for answering ES|QL related questions.

The utility function (executeAsReasoningAgent) will be used in Streams for various LLM-based workflows.

…bana into inference-prompt-reasoning

dgieselaar · 2025-07-30T05:44:06Z

@elasticmachine merge upstream

qn895 · 2025-07-30T18:24:52Z

x-pack/platform/packages/shared/kbn-ai-tools/src/tools/describe_dataset/index.ts

+      fields: '*',
+      index_filter: {
+        bool: {
+          filter: rangeQuery(start, end),


Wonder if it would be beneficial to restrict this query to certain data tiers, or exclude frozen tiers here.

the rangeQuery will do that, based on time. It will only hit frozen/cold/warm if the data is within range - ES will pre-filter shards for data streams based on time range. It might still hit frozen if it queries a wide time range, but that's maybe more of a feature - otherwise it cannot query frozen even if the user asks for it.

x-pack/platform/packages/shared/kbn-ai-tools/src/tools/esql/index.ts

qn895 · 2025-07-30T19:44:14Z

...packages/shared/kbn-ai-tools/src/tools/describe_dataset/sort_and_truncate_analyzed_fields.ts

+  const { fields, ...meta } = analysis;
+  const [nonEmptyFields, emptyFields] = partition(analysis.fields, (field) => !field.empty);
+
+  const sortedFields = [...shuffle(nonEmptyFields), ...shuffle(emptyFields)];


Can you add a comment to note why we are calling shuffle() here?

yes! it's to get a random-ish subset of fields, so you don't end up with a-f and drop everything on the floor. (i'll add this as a comment)

x-pack/platform/packages/shared/kbn-ai-tools/src/tools/esql/esql_task_description.text

…ql_task_description.text Co-authored-by: Quynh Nguyen (Quinn) <43350163+qn895@users.noreply.github.com>

qn895

LGTM as we can address the remaining comments in the follow up 🎉

dgieselaar · 2025-08-01T11:26:18Z

@elasticmachine merge upstream

elasticmachine · 2025-08-01T13:16:16Z

💚 Build Succeeded

Buildkite Build
Commit: 3e0cbfd

Metrics [docs]

Public APIs missing comments

Total count of every public API that lacks a comment. Target amount is 0. Run node scripts/build_api_docs --plugin [yourplugin] --stats comments for more detailed information.

id	before	after	diff
`@kbn/ai-tools`	-	23	+23
`@kbn/inference-common`	171	173	+2
`@kbn/inference-prompt-utils`	-	4	+4
total			+29

Public APIs missing exports

Total count of every type that is part of your API that should be exported but is not. This will cause broken links in the API documentation system. Target amount is 0. Run node scripts/build_api_docs --plugin [yourplugin] --stats exports for more detailed information.

id	before	after	diff
`@kbn/ai-tools`	-	2	+2

Unknown metric groups

API count

id	before	after	diff
`@kbn/ai-tools`	-	23	+23
`@kbn/inference-common`	318	320	+2
`@kbn/inference-prompt-utils`	-	4	+4
total			+29

History

💛 Build #324987 was flaky 733616d
💛 Build #324568 was flaky 120a3e5
💔 Build #324491 failed eccc031
💔 Build #324274 failed ea79a80

Adds a utility function that will execute a Prompt (system message, user message template and tool definitions), in a way that the LLM will reason about the task and its progress and execute tools autonomously until it feels confident enough to answer the question. As an example, a reference implementation was added for answering ES|QL related questions. The utility function (`executeAsReasoningAgent`) will be used in Streams for various LLM-based workflows. --------- Co-authored-by: kibanamachine <42973632+kibanamachine@users.noreply.github.com> Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com> Co-authored-by: Quynh Nguyen (Quinn) <43350163+qn895@users.noreply.github.com>

kibanamachine · 2025-08-05T13:48:47Z