Release AI Catalog sub-agent references as a beta
What does this MR do and why?
Releases AI Catalog sub-agent references (include in a custom flow) as a beta.
- The
ai_catalog_flow_agent_referencesflag moves fromwiptobeta. The flag stays off by default, and the rollout on GitLab.com uses chatops. - The security review in #632224 must be complete before the flag is on by default.
ai_catalog_flow_agent_referencesis inGitlab::Llm::Utils::AiFeaturesCataloguewithmaturity: :beta, andGitlab::Llm::StageCheckmaps it to theai_cataloglicense.- A flow with
includesaves only whenStageCheck.available?is true for the project. For a beta feature, that needs the instance setting for experiment and beta features and, on GitLab.com, the same setting on the top-level group. - The
catalog_itemsendpoint uses the same check before it resolves sub-agents. - Flows without
includedo not change. - The docs mark
includeas a beta with a flag that is disabled by default.
Merge order
Blocked by !259499 (merged), because the docs change here edits the include section that !259499 (merged) adds. This MR targets the !259499 (merged) branch. When !259499 (merged) merges, GitLab changes the target to master.
References
- Issue: #632225 (closed)
- Rollout issue: #632224
- Epic: &24006 (closed)
- Decision to use the experiment and beta features setting as the opt-in: gitlab-org/modelops/applied-ml/code-suggestions/ai-assist#2965 (comment 3937064271)
- Workspace agents MR with the same shape: !259664 (merged)
Screenshots or screen recordings
Not applicable. Backend only.
How to set up and validate locally
-
Turn the flag on:
Feature.enable(:ai_catalog_flow_agent_references) -
Create an agent in the AI Catalog, release version
1.0.0, and note its item ID. -
Turn the setting off:
ApplicationSetting.current.update!(instance_level_ai_beta_features_enabled: false) -
Create a custom flow in the same project and add this to the definition:
include: - source: ai-catalog item_type: agent item_id: "<id>" version: "1.0.0" -
The save fails with "Referencing AI Catalog items with include needs experiment and beta features turned on".
-
Turn the setting on (
truein step 3) and save again. The save succeeds. -
With the setting off, save a flow without
include. The save succeeds.
MR acceptance checklist
Evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.