Make convex
Searchable.

The Sync Engine & Fast Bootstrap

When you first connect a table with millions of rows, syncing them via individual events would take hours. Instead, the Lytibox daemon hooks directly into Convex's snapshot export system.

It requests a native snapshot, downloads the resulting .zip, and bulk-indexes the raw data into Manticore at blazing speeds. Once the baseline is established, it seamlessly transitions to polling the lytibox_events table for sub-second, real-time incremental updates. It efficiently batches insert, replace, and delete operations.

Dynamic Schema Evolution

Lytibox's schema management is completely decoupled from your Convex database schema. You can map arbitrary, dynamic, or user-defined custom fields directly into your search targets during the materialization phase.

When the daemon encounters a materialized document containing a previously unseen field, it dynamically infers the correct Manticore data type (float, multi64, text, etc.) straight from the JSON payload. It then automatically executes an ALTER TABLE command against the search engine before inserting the row. This means if your users create custom attributes in your app, those properties instantly become queryable, filterable, and sortable columns in Manticore, with no manual migrations required.

Native Types and Text Search

Manticore is a phenomenal C++ search engine. We don't just dump JSON into a text field. We map Convex data directly to Manticore's strongly-typed columns: integer, float, boolean (mapped efficiently), string, timestamp (for dates), and multi64 arrays.

Because the index is strictly typed, exact-match filters are computationally cheap. Simultaneously, Full-Text Search (FTS) runs in parallel on dedicated text fields using advanced tokenization, allowing you to build rich multi-field search bars that rank by relevance out of the box.

Expressive AST & The Operator Arsenal

Our React hook exposes an Abstract Syntax Tree (AST) that compiles down to highly optimized Manticore SQL on the server. You are not limited to simple equality checks. You can build infinitely nested AND / OR groups, paired with a massive standard library of operators:

Whether you are querying integer ranges, checking if an array HAS ALL required tags, or filtering dates IN LAST 7d, the compiler handles the exact translation logic safely.

Relational Queries Without Joins

Convex naturally handles relational data via Document IDs. Lytibox allows you to resolve and denormalize these during the materialization phase. For example, when indexing an order, you can fetch the associated customer's name and embed it directly into the Manticore search index. This allows you to filter and run text searches on nested relational data instantly, without relying on slow joins at query time.

Aggregations, Stats, and Suggestions

Because the query engine operates independently from the transactional database, we don't just return flat rows. A single query can simultaneously return:

• Hydration IDs: The paginated Convex Document IDs to pass back to your UI.
• Total Counts: The absolute match count of the filter, bypassing pagination limits.
• FTS Suggestions: Dynamic auto-complete for your inputs, returning exact occurrence counts for a given keyword within the currently filtered dataset.
• Stats & Graphs: Grouped statistical data (counts, sums, averages) perfect for rendering bar charts and complex reporting dashboards natively alongside your tables.

Sorting & Pagination

Every target supports sorting across multiple columns (ASC and DESC) with automatic stable fallback sorting to prevent pagination drift. The UI handles the cursors and offsets automatically, so building a massive, heavily filtered, paginated admin table takes minutes, not days.