[SigEvents] Remove custom model selection, migrate index patterns to uiSettings

[achyutjhunjhunwala]

Summary

Cleans up the Significant Events settings page by removing our own per-feature model selection infrastructure and delegating connector resolution to the platform's Inference Feature Registry.

Before: The Settings tab had 3 connector dropdowns (KI extraction, query generation, discovery). Selections were stored in a custom saved object (streams-significant-events-settings). Background tasks read that SO to decide which LLM to call.

After: The Settings tab has a link to Stack Management → Model Settings, where admins configure connector overrides centrally (same place as every other inference feature in Kibana). The index patterns field stays on the page but is now stored as a Kibana uiSetting instead of the custom SO. The custom SO is gone entirely.

---

What changed and why

Settings tab (tab.tsx)

The 3 connector dropdowns, their useLoadConnectors hooks, stale-connector callouts, and no-default callouts are all removed. In their place: a short paragraph and an EuiLink to the Model Settings page, generated via MANAGEMENT_APP_LOCATOR (the same locator pattern used by SLO, Synthetics, etc.).

The index patterns textarea stays. It now reads the initial value from core.uiSettings.get() and saves via core.settings.client.set() — the built-in uiSettings HTTP API handles persistence, no custom route needed.

New uiSetting: observability:streamsSigEventsIndexPatterns

• Registered in feature_flags.ts alongside the other streams flags
• readonly: true + readonlyMode: 'ui' — hidden from the Advanced Settings page, but the custom settings page can still read and write it programmatically
• Default value: logs* (same as the old SO default)
• Per-space (namespace scope), consistent with how other streams settings work

Task definitions — connector resolution

All 4 tasks (features_identification, significant_events_queries_generation, insights_discovery, onboarding) previously read connectorId* fields from the custom SO and passed them through resolveConnectorIdAndCheckAllowlist. They now call getForFeature directly:

if (!taskContext.server.searchInferenceEndpoints) {
  throw new StatusError('Inference endpoints plugin is unavailable.', 503);
}
const { endpoints } = await taskContext.server.searchInferenceEndpoints
  .endpoints.getForFeature(FEATURE_ID, fakeRequest);
if (endpoints.length === 0) {
  throw new StatusError('No connector configured. Configure one in Model Settings.', 400);
}
const connectorId = endpoints[0].connectorId;

getForFeature resolves the connector using this priority order:

1. Admin override set via the Model Settings page (stored in inference_settings SO)
2. recommendedEndpoints from our feature registration (register_significant_events_inference_features.ts)
3. Platform default connector

useIndexPatternsConfig hook

Replaced the API fetch (GET /internal/streams/_significant_events/settings) with a direct core.uiSettings.get() read. Return signature is unchanged so the only consumer (streams_view.tsx) needs no updates.

---

routes/utils/resolve_connector_id.ts is kept — it is still used by the significant events SSE endpoint and description generation route, which accept a caller-supplied connectorId with a fallback to the default AI connector.

---

Test updates

Core SO integration tests — two platform-level tests that track all registered SO types were updated to remove streams-significant-events-settings:

• src/core/server/integration_tests/saved_objects/registration/type_registrations.test.ts
• src/core/server/integration_tests/ci_checks/saved_objects/check_registered_types.test.ts

Evals suite — configureModelSelectionSettings() in kbn-evals-suite-significant-events previously called PUT /internal/streams/_significant_events/settings (now deleted). Updated to call PUT /internal/search_inference_endpoints/settings instead — the platform admin override API that getForFeature checks first.

---

Data migration

Users with custom index patterns saved in the old SO will have them reset to logs* after this change. Acceptable given the feature is in technical preview with limited adoption. No migration script — the old orphaned SO data causes no errors (Kibana ignores unknown SO types).

---

How to test

1. Go to Significant Events Discovery → Settings tab
2. Verify the LLM section shows descriptive text and a Go to Model Settings link (opens Stack Management → Model Settings)
3. Verify the index patterns field is still editable; save and confirm it persists across page reload
4. In Stack Management → Model Settings, set a connector override for one of the streams sig events features
5. Trigger a features identification task and check Kibana logs for:

   Using connector <id> for knowledge indicator extraction
   

   (Requires logging.loggers[plugins.streams].level: debug in kibana.dev.yml)
6. Confirm the task used the connector configured in step 4

New Custom Settings Page

[achyutjhunjhunwala]

/ci

[ruflin]

Did manual testing of the PR, works as expected. When the default model is changed, a different model is used. Otherwise, all the default models are used.

[achyutjhunjhunwala]

/ci-ralph

[achyutjhunjhunwala]

/ralph fix the build issues and make it green

[achyutjhunjhunwala]

Fixed the CI build failures from build #420664:

1. SAVED_OBJECT_TYPES_COUNT (Checks failure): Decremented from 155 to 154 in src/core/packages/saved-objects/server-internal/src/object_types/index.ts to reflect the removal of the streams-significant-events-settings SO type.
2. tsconfig.json (Quick Checks - ts_projects): Updated streams_app/tsconfig.json — removed unused @kbn/inference-connectors reference, added required @kbn/deeplinks-management (used by the new Management Settings link).
3. moon.yml (Quick Checks - verify_moon_projects): Regenerated streams_app/moon.yml to match the tsconfig changes.

Verified the saved objects registration integration test passes locally.

Written by cursor, wait for @achyutjhunjhunwala to chime in

[tonyghiani]

LGTM 👌

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR is a significant architectural change that removes a saved object type and migrates connector/model selection to use the centralized Inference Feature Registry. While mostly deletions, it changes runtime behavior for connector resolution and removes settings API endpoints. The author does not own any of the changed files, which are owned by multiple teams including obs-sig-events-team and kibana-core.

^{You can customize Macroscope's approvability policy. Learn more.}

[achyutjhunjhunwala]

/ralph check the test failures of build 421192 and fix them

Build URL: https://buildkite.com/elastic/kibana-pull-request/builds/421192

Use ci/bk-build-info.sh 421192 to fetch detailed build failure information.

[achyutjhunjhunwala]

Investigated the test failures in build #421192. All 3 failures are not caused by this PR:

1. LLM Evals (2 steps — eis-anthropic-claude-4-6-sonnet and eis-openai-gpt-oss-120b): Pre-existing bug in kbn-scout-reporting — the getKibanaModulePath function in src/platform/packages/private/kbn-scout-reporting/src/helpers/read_manifest.ts requires a scout or scout_* directory in the playwright config path, but evals packages place playwright.config.ts at the package root. This was introduced by commit 8741ad7e6140 merged to main and affects all evals suites, not just significant-events.

2. FTR Configs Issue #60 #63: Unrelated flaky UI test ("search features index details page").

Previous builds #421000 and #420993 passed (with flaky retries), confirming these are not PR-related. No action needed in this PR.

Written by Ralph, wait for @achyutjhunjhunwala to chime in

[leathekd]

Schema LGTM

[mattkime]

codeowner changes lgtm

[elasticmachine]

💛 Build succeeded, but was flaky

• Buildkite Build
• Commit: f90c53c

Failed CI Steps

• FTR Configs #2
• FTR Configs #23
• FTR Configs #24
• FTR Configs #26
• FTR Configs #44
• FTR Configs #51
• FTR Configs #64
• FTR Configs #79
• FTR Configs #90
• FTR Configs #93
• FTR Configs #118
• FTR Configs #125
• FTR Configs #138
• FTR Configs #148
• FTR Configs #149
• FTR Configs #165
• FTR Configs #167
• FTR Configs #171
• FTR Configs #175
• FTR Configs #176
• FTR Configs #178
• FTR Configs #179
• FTR Configs #184
• FTR Configs #188
• FTR Configs #189
• FTR Configs #197

Metrics [docs]

Module Count

Fewer modules leads to a faster build time

id	before	after	diff
streamsApp	1782	1739	-43

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/management-settings-ids	148	149	+1

Async chunks

Total size of all lazy-loaded chunks that will be downloaded as the user navigates the app

id	before	after	diff
streamsApp	1.9MB	1.8MB	-11.5KB

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
streams	30	29	-1

Unknown metric groups

API count

id	before	after	diff
@kbn/management-settings-ids	149	150	+1

History

• 💔 Build #421192 failed 26575bb
• 💛 Build #421000 was flaky 4d78c45
• 💛 Build #420993 was flaky 4d78c45
• 💔 Build #420920 failed 9e80af3
• 💔 Build #420875 failed f02e290

cc @achyutjhunjhunwala