AI Search (Semantic Search)

AI search adds an embedding-based retrieval lane alongside BM25. Long-form content stays retrievable when the wording differs from the query, and hybrid mode combines both lanes for the best of each.

1. Pick the fields to embed

1. Settings → Search → AI Search → + Add more.
2. Pick the (node type, field) pairs to embed.
3. For each field, decide on chunking (below).

Good defaults:

Don't embed identifiers — they have no semantic content and you'll burn embedding budget for nothing.

2. Configure chunking

For anything longer than the embedding model's context window, enable Chunk Text. The workspace splits the field into overlapping windows and embeds each one independently. Hits return the parent node, deduplicated.

Knobs:

3. Similarity cutoffs

Two cutoffs control how strict matching is:

• Added cutoff — minimum cosine similarity for an embedding hit to enter the result set.
• Rerank cutoff — minimum similarity for an item to re-rank an existing BM25 hit.

Defaults are usually fine. Tune up if the LLM hallucinates over weak retrievals (raise both); tune down if relevant items are missing (lower both).

4. Semantic reranker

The reranker is a second pass that uses a cross-encoder model to re-score the top-N hits from hybrid retrieval. Enable when:

• Precision@10 matters more than throughput.
• Your corpus is large enough that BM25 + embeddings still return some near-misses.

It's an extra call per query; budget accordingly.

From code

var req = SearchRequest.For("laptop overheating");
req.BeforeTypesFacet = new HashSet<string> { N.SupportCase.Type };
req.HybridSearch     = true;          // Use BM25 + embeddings together
req.SemanticRerank   = true;          // Apply the cross-encoder reranker

var query = await Graph.CreateSearchAsUserAsync(req, CurrentUser, CancellationToken);
return query.Take(10).EmitWithScores();

For pure semantic retrieval (no BM25), use:

return (await Q().StartAtSimilarTextAsync("laptop overheating",
                                          nodeTypes: new[] { N.SupportCase.Type }))
          .Take(10)
          .EmitWithScores();

Budget and performance

• Embedding cost scales with the field corpus size on first index, then with the change rate.
• Pause embedding generation under Settings → Search Index before a large backfill; resume after.
• Embedding latency on query is fixed and small (~10 ms). Hybrid mode is fast; reranking adds 50–200 ms.

LLM Search: what the assistant may search

The settings above configure the retrieval lane. A separate tab in the same place — Settings → Search → LLM Search, next to AI Search — configures what the AI assistant's built-in search tool is allowed to reach. The two are independent: embedding a field does not expose it to the assistant, and exposing a data type to the assistant does not embed anything.

The page has two tabs:

• Data Types — one row per node type, toggled between Searchable by AI and Not searchable by AI. This is a ceiling, not a grant: a user still only searches the types they can see.
• Filters — one row per filter the interpreter may apply, toggled between Available to AI and Not available to AI. Each row shows the name the model sees in bold, what kind of filter it is (field value, date range, data type, related item), and the back-end facet key behind it. Which filters are listed depends on the data types enabled in the first tab, and turning a data type off drops its filters rather than leaving them as dead configuration.

While nothing is configured, the built-in defaults apply: the _FileEntry and _WebPage data types, and the file-kind, file-extension, source and time filters.

Both lists are part of the general configuration, so they travel with configuration export and import. A configuration file written before the setting existed omits them, and the importer leaves the workspace as configured; an explicitly empty list imports as "back on the built-in defaults".

Cross-links

• Full text search
• Ranking and boosting
• Relevance tuning
• The search tool
• Embeddings (concepts)
• Vector search reference
• Hybrid search reference