AddThisFeature

Algolia Index Sync

Keep the search index matching the database, including who is allowed to see what.

involved Integrations

What it adds

A sync pipeline that mirrors records into a search index with tenant-scoped filters and scheduled reconciliation.

What your agent is told to do

5
  1. 1

    Identify every record type that should be searchable and define one index document per record, keyed by an identifier that never changes for the life of that record.

  2. 2

    Push changes through the app's existing background-job system on create, update, and delete, and make every write idempotent so a replayed job produces the same document rather than a duplicate.

  3. 3

    Put tenant, workspace, and visibility values on each document as filter attributes, and have the server issue short-lived scoped search credentials that pin those filters. The browser must never choose its own filter.

  4. 4

    Run backfills in batches sized to the provider's limits and throttled so a reindex does not starve live updates, and make a full rebuild something an operator can run without taking search offline.

  5. 5

    Schedule a reconciliation that compares the index against the source of truth and repairs the difference. Search indexes drift, and a system with no reconciliation will be quietly wrong within a month.

Edge cases it handles

8
  • An identifier derived from mutable data such as a slug or a title creates a duplicate document on every rename. Key documents by the record's permanent identifier.
  • A deleted record, or one whose access was revoked, must leave search immediately. A stale document that still matches is a data leak, not a cosmetic bug.
  • Filters enforced only in client code can be edited by the user. The filter has to be signed into the search credential the server issues, and that credential has to expire.
  • Bulk operations exceed the provider's request size and rate limits. Chunk the payload, respect the retry interval it returns, and back off rather than dropping records on the floor.
  • Indexing jobs fail, notifications are missed, and deploys interrupt queues. Reconciliation must find both missing and orphaned documents and repair each without a manual export.
  • Index writes are eventually consistent. A user who saves a record and searches for it a second later may not find it, so drive the immediate view from the database rather than from search.
  • Indexed content can include fields the searching user may not see. Decide per field what is indexed and returned, and keep private fields out of the document entirely rather than hiding them at render time.
  • When the provider is unavailable the app must fall back to whatever database-backed search or listing it already has, with a notice, rather than showing an empty result page.

Definition of done

9
  • Every searchable record type has a document keyed by a permanent identifier.
  • Create, update, and delete propagate through background jobs and are idempotent on replay.
  • Deletion and access revocation remove the record from search results.
  • Search credentials are issued server-side, scoped to the caller's tenant and visibility, and expire.
  • Backfills batch within provider limits and do not block live updates.
  • A scheduled reconciliation finds and repairs missing and orphaned documents.
  • A provider outage falls back to the app's own search with a visible notice.
  • The feature matches the existing design system.
  • No existing functionality is broken.

Related features

How it works

  1. 1

    Copy the link

    Grab the Markdown instruction URL for this feature.

  2. 2

    Give it to your AI

    Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.

  3. 3

    It inspects, then implements

    Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.