activate_pipeline
ChatGPTActivate a ranking pipeline so it begins serving ranked results.
Overview
Inspect and manage your NeuronSearchLab recommendation workspace from ChatGPT. Search catalogue items, understand rankings, configure contexts, pipelines, rules and segments, run experiments and model training, and review performance analytics.
Tools
cancel_training_job
ChatGPTCancel a running training job.
clone_pipeline
ChatGPTDuplicate an existing pipeline with all its stages and configuration. Useful for creating experiment variants or safe copies before making changes.
compare_items
ChatGPTCompare two items head-to-head by served count, events, clicks, and click-through rate over the same time window.
create_context
ChatGPTCreate a new recommendation context (feed). A context defines a feed surface — e.g. a homepage feed, a "you may also like" section, or a search assist panel. After creating a context, create a pipeline and optionally rules to control ranking.
create_event_type
ChatGPTCreate a new event type that can be tracked in the recommendation model. The value (0–100) controls how much weight this signal has during training.
create_experiment
ChatGPTCreate a new A/B experiment comparing pipeline variants. Specify 2+ variants with traffic fractions summing to 1.0. Each variant can point to a different pipeline_id. Start it with start_experiment after creating.
create_pipeline
ChatGPTCreate a new ranking pipeline with default stages. Optionally attach it to a context. The pipeline starts active by default.
create_rule
ChatGPTCreate a new ranking rule. Rule types: boost (increase score), bury (decrease score), pin (fix position), filter (remove items), cap (limit fraction), diversity (spread categories). Each rule has conditions (which items match) and an action (what to do).
create_segment
ChatGPTCreate a new user segment with membership conditions. Condition types: behavioral (purchase/click history), demographic (age, location), computed (ML scores), item_interaction (affinity to categories or items). Example: power users who purchased 3+ times in 30 days.
create_training_job
ChatGPTStart a model training job from an existing template. This consumes training quota and may incur usage charges; auto-promotion options can replace the model serving live traffic.
deactivate_pipeline
ChatGPTDeactivate a ranking pipeline so it stops serving results without being deleted.
delete_context
ChatGPTDelete a recommendation context. This also removes associated pipelines and rules.
delete_event_type
ChatGPTPermanently delete an event type. This cannot be undone. Existing tracked events referencing this type will remain in the database but the type definition will be removed.
delete_pipeline
ChatGPTDelete a ranking pipeline.
delete_rule
ChatGPTDelete a ranking rule.
delete_segment
ChatGPTDelete a user segment. Any rules referencing this segment will need to be updated.
disable_rule
ChatGPTDisable a ranking rule without deleting it. Useful for temporarily pausing rules during debugging or campaigns.
enable_rule
ChatGPTEnable a ranking rule that was previously disabled.
explain_ranking
ChatGPTExplain why a specific item was (or would be) ranked at a particular position for a user. Returns a score breakdown showing embedding similarity and applied rules, plus a pipeline trace showing which stages passed or failed.
get_account_plan
ChatGPTGet the authenticated team's current billing plan, resolved resource limits, live resource counts, metered usage, and exact overages. Always call this first when the user asks to clean up plan-limit overages, reduce usage to plan limits, or check remaining capacity.
get_auto_recommendations
ChatGPTFetch one auto-generated recommendation section for a user through the hosted console bridge. Serving a section records attribution and metering data and preserves section, next_cursor, and done response metadata.
get_context
ChatGPTGet the full configuration of a specific recommendation context by ID.
get_demo_recommendations
ChatGPTUse after try_demo or record_demo_interaction to fetch a fresh recommendation result for a reserved synthetic user. This is only for the NSL sample workspace and never reads or changes a real customer user.
get_experiment
ChatGPTGet the full configuration of a specific A/B experiment including variants, traffic splits, and current metrics.
get_experiment_results
ChatGPTGet the most recently calculated statistical results for a completed or running experiment without changing stored data.
get_item_analytics
ChatGPTGet served counts, event breakdown, watch/click counts, and click-through rate for a specific item over a time window.
get_onboarding_status
ChatGPTPoll after connect_my_site. Reports how many pages were read, how many became catalogue items, and how many were skipped. Report the coverage numbers to the user rather than only the item count - a catalogue built from pages with no structured data ranks poorly, and they should know which they have.
get_pipeline
ChatGPTGet the full configuration of a specific ranking pipeline by ID.
get_ranking_metrics
ChatGPTGet ranking performance metrics for a pipeline or context: CTR by position, overall CTR, conversion rate, diversity score, catalogue coverage, and revenue per session.
get_recommendations
ChatGPTFetch personalised recommendations for a user through the hosted console bridge. Serving a result records attribution and metering data and returns a request_id for subsequent events.
get_rule
ChatGPTGet the full configuration of a specific ranking rule by ID, including conditions and action.
get_segment
ChatGPTGet the full definition of a specific user segment, including all its conditions.
get_training_job
ChatGPTGet a customer-facing status summary for a specific training job without returning infrastructure identifiers or raw logs.
get_user_analytics
ChatGPTGet served counts, event breakdown, unique-item activity, and click-through rate for a specific user over a time window.
list_api_keys
ChatGPTList all API keys for the team with their names, scopes, environments, and last-used timestamps.
list_contexts
ChatGPTList all recommendation contexts (feeds) configured for your team. Contexts define how recommendations are scoped, filtered, and grouped.
list_event_types
ChatGPTList all configured event types (click, view, purchase, etc.) for the team. Returns each event type with its ID, name, weight/value (0–100), and recorded event count.
list_experiments
ChatGPTList all A/B experiments with their current status (draft, running, paused, completed).
list_integrations
ChatGPTList configured third-party integrations (webhooks, data connectors, export targets, etc.).
list_pipelines
ChatGPTList all ranking pipelines. Pipelines define the sequence of stages (candidate generation, scoring, rules, post-processing) that transform raw candidates into a ranked feed.
list_rules
ChatGPTList ranking rules, optionally filtered by context. Rules modify the ranking output — boost, bury, pin, filter, cap, or diversify items.
list_segments
ChatGPTList all user segments. Segments group users by shared behaviours or attributes and can be referenced in rules and campaigns.
list_training_jobs
ChatGPTList recent model training jobs with their status (running, completed, failed). Training jobs re-train the ranking model on new interaction data.
record_demo_interaction
ChatGPTRecord one synthetic view, click, or purchase for a reserved demo user and sample item. Use this only after showing the first demo recommendations, then call get_demo_recommendations so the user can see how the result changes. The operation is idempotent for a retried MCP request.
refresh_experiment_results
ChatGPTRecalculate experiment metrics from current recommendation and event data, store the refreshed metrics, and return the updated experiment results.
render_demo_recommendations
ChatGPTPresentation-only tool. Call try_demo first, wait for stage ready, then pass its complete structured result here unchanged. Renders the synthetic user, recent activity, ranked recommendations, and safe interactive sample controls in ChatGPT. Do not call this for real customer data or a pending workspace.
revoke_api_key
ChatGPTRevoke an API key. This immediately invalidates the key and cannot be undone.
search_items
ChatGPTSearch the recommendation catalogue by keyword. Returns matching items with their IDs, names, descriptions, and status. Useful for looking up specific items before updating or removing them.
start_experiment
ChatGPTStart an A/B experiment. Traffic will begin splitting between variants immediately. The experiment moves from draft to running status.
start_using_my_data
ChatGPTUse after the user has experienced the demo and wants to move toward their own catalogue. Returns honest, supported handoff choices for a feed or CSV import, website integration, or code integration. It does not create credentials or claim that real events are connected.
stop_experiment
ChatGPTStop a running A/B experiment and mark it as completed. Results are preserved for analysis.
toggle_rule
ChatGPTEnable or disable a ranking rule without deleting it.
top_items
ChatGPTList the top items by served count or by matching event activity such as watch or click over a time window. For generic "top item" or "best performing item" questions, prefer metric="served" unless the user explicitly names an event type.
try_demo
ChatGPTUse this first when the user wants to try NeuronSearchLab, asks what it can do, or requests recommendations without naming production data. It prepares a safe synthetic workspace and, once ready, returns a demo user plus the first real NSL recommendation result. Do not ask the user to configure NSL before calling it.
update_context
ChatGPTUpdate an existing recommendation context.
update_event_type
ChatGPTUpdate an existing event type's name or weight. Use list_event_types first to find the event_id.
update_experiment
ChatGPTUpdate an experiment that is in draft status.
update_pipeline
ChatGPTUpdate an existing ranking pipeline.
update_rule
ChatGPTUpdate an existing ranking rule.
update_segment
ChatGPTUpdate an existing user segment definition.
Capabilities
WritesInteractive
Example Prompts
Click any prompt to copy it.
