Skip to main content
A context is a named configuration that shapes what the recommendation engine returns for a specific surface. You create one context per placement, such as a homepage feed, a product detail page, or an email digest, and reference its ID in every API call for that surface. This guide covers all the controls available inside the context editor.

​
Create a context

  1. Open Console > Contexts and click New context.
  2. Give it a descriptive name such as “Homepage Feed” or “Post-purchase Sidebar”.
  3. Add an optional description so teammates understand its purpose.
  4. Select the model this context draws recommendations from.
  5. Choose the recommendation type:
    • Item to User returns items relevant to a given user.
    • User to Item returns users similar to a given item.
    • Item to Item returns items similar to a given item.
    • User to User returns users similar to a given user.
  6. Save the context and note its numeric ID. Your engineering team will pass this ID in context_id on every recommendation request.

​
Filters

Filters restrict which items are eligible for recommendation. Only items that pass all active filters will appear in results. To add a filter:
  1. In the Filters section, click Add Condition.
  2. Select a metadata field from your catalog (for example category, in_stock, price).
  3. Choose an operator: =, !=, LIKE, NOT LIKE, >, <, >=, <=, IN, NOT IN.
  4. Enter the value to compare against.
  5. Add more conditions and set each one to combine with the previous using AND or OR.
Example: To restrict a context to in-stock items in the “Electronics” category:
FieldOperatorValueLogic
in_stock=true
category=ElectronicsAND

​
Influence rules

Influence rules adjust the score of items that match a condition, moving them up or down relative to other results without hard-filtering them out. To add an influence rule:
  1. In the Influence section, click Add Influence Rule.
  2. Set the type to Promote (score higher) or Demote (score lower).
  3. Select the field to match on (for example category or a specific item_id).
  4. Enter the value to match.
  5. Choose the weight:
    • +10%, +25%, +50%, 2x for promotion
    • -25%, -50% for demotion
Example: To promote new arrivals while keeping older items in the mix:
TypeFieldValueWeight
Promoteis_new_arrivaltrue+50%

​
Group by fields

Group by fields enforce diversity by limiting how many items from the same group can appear consecutively or within a result set. To add a group by field:
  1. In the Group By Fields section, click Add Group By Field.
  2. Select the metadata field to group on, such as category or brand.
When grouping is active, the engine will spread results across distinct values of that field rather than surfacing many items from the same group.

​
Sliding window for repeat items

The sliding window prevents items a user has recently seen from appearing again too soon.
  1. Check Exclude items the user has already viewed.
  2. Set the Amount and Unit (Minutes, Hours, Days, Weeks, or All time).
Setting the unit to All time permanently excludes any item the user has ever viewed. Choosing a time window such as 7 Days means items reappear after that period.

​
Pinned and forced items

Use Ensure top X to guarantee that specific items always appear within the first results.
  1. Check Ensure top X in the additional options.
  2. Set Top X count to the number of positions to fill with forced items.
  3. Specify the Forced items by item ID.
This is useful for sponsored placements, editorial picks, or legally required disclosures that must always be surfaced.

​
Random content injection

Checking Include random content in recommendations mixes a small proportion of randomly selected items into results. This introduces serendipity and helps the engine discover new interests for users who have a narrow history.

​
Preview recommendations

Use the Preview Recommendations panel at the bottom of the context editor to test the current configuration before saving.
  1. Select a user from the dropdown, or click Random User to pick one at random.
  2. The panel displays the items that would be returned for that user under the current context settings.
Preview results are live: changes to filters, influence rules, or other settings update the preview immediately, letting you verify the output before committing.

​
Iterate over time

As your product evolves, return to the context editor to adjust settings. Changing influence weights, adding new filters, or widening the repeat-item window takes effect immediately for all subsequent API calls using that context ID. No redeployment is required.