Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
A search method or type, like full text, vector, or hybrid, controls how Agents and Tools with Foundry Local retrieves and ranks results from your indexed data. It shapes the way users interact with your solution and the quality of the answers they get.
You select a search type when you configure your data query settings in the Agents and Tools with Foundry Local developer portal. Text, vector, and hybrid search are available in all deployments. Hybrid multimodal search requires GPU mode (gpu_enabled=true so image embedding collections are provisioned) and, in the developer portal, a configured byom.apiEndpoint with a vision-capable model.
Why your search type matters
Your choice of search type affects:
- The kinds of questions users can ask.
For example: Text search works best for simple keyword queries, while vector or hybrid search lets users ask more natural, conversational questions. - The relevance and accuracy of results. For example: Hybrid search can return more relevant answers by combining exact matches with semantic understanding, while text search might miss contextually similar results.
- Access to advanced features like multimodal search. For example: Hybrid multimodal search lets users find information in both text and images, which is a capability not available with other search types.
Available search types
The following definitions describe each search type available in Agents and Tools with Foundry Local, so you can understand how they work and choose the best option for your scenario.
Full text search: A search method that scans and matches the entire body of text in documents. Use keywords, phrases, or boolean queries to find relevant chunks in the supplied documents.
Hybrid search: Combines both full-text search (keyword-based) and vector search (semantic similarity) to retrieve the most relevant documents. It uses the precision of keyword matching and the depth of semantic understanding for improved retrieval accuracy.
Hybrid multimodal search: Combines full-text and vector search across both text and image data. Agents and Tools with Foundry Local can retrieve and rank results from multiple data types in a single query. Unlike hybrid search, which works only with text, by using both keyword and semantic similarity, hybrid multimodal search extends this approach to include images. This approach enables richer, more comprehensive answers that draw from both textual and visual content.
Vector search: A search method that finds relevant documents by comparing the semantic similarity between vector embeddings of the user's query and precomputed embeddings of documents. This method typically uses cosine similarity or other distance metrics in a vector space.
Compare search types
The Agents and Tools with Foundry Local extension supports several search types. The following table summarizes each search type, best use cases, and performance considerations:
| Search type | What it does | Best use cases | Performance notes |
|---|---|---|---|
Text search (full_text_search) |
Scans and matches exact words or phrases in your documents by using keywords, phrases, or boolean queries. | Standard question and answer (Q&A), document lookup, legal, or compliance scenarios | Fastest for keyword lookups; might miss semantic matches. |
| Vector search | Uses semantic similarity to find contextually relevant results by comparing vector embeddings. | Conversational Q&A, finding related content | Fast; slightly more compute than text search. |
| Hybrid search | Combines text (keyword) and vector (semantic) search for better coverage and accuracy. | Most scenarios; default for general use | Fast; combines strengths of text and vector search. |
| Hybrid multimodal | Combines text, vector, and image search to retrieve and rank results from multiple data types in a single query. | Scenarios involving both text and images | Slower than text/vector; performance depends on data size and type. |
Search type availability by deployment
Text, vector, and hybrid search are available in all deployments. Hybrid multimodal search requires GPU mode (gpu_enabled=true so image embedding collections are provisioned) and, in the developer portal, a configured byom.apiEndpoint with a vision-capable model:
| Search type | Available |
|---|---|
| Hybrid search | Yes |
| Text search | Yes |
| Vector search | Yes |
| Hybrid multimodal search | Requires GPU mode (gpu_enabled=true) and, in the developer portal, a configured byom.apiEndpoint with a vision-capable model |
For more information about deploying with a language model, see:
- Deployment overview for Agents and Tools with Foundry Local
- Deploy the extension for Agents and Tools with Foundry Local
How to select a search type
Select the search type in the Agents and Tools with Foundry Local developer portal when you configure your data query settings. Some parameters might differ by search type and model.
Collection-scoped search
You can scope all search types to specific collections. When you query through the developer portal or API, specify which collections to search:
- Developer portal: Queries the default
edgeragappcollection only. The portal doesn't offer a collection picker. - Inference API: External
POST /chat/completionsis deprecated and restricted to the developer portal (which queriesedgeragapp). External or agentic queries over collections use anindexed_sources_mcpknowledge source via/openai/v1/responsesor/mcp. - MCP Server: Set
collection_namesin the tool arguments on/mcp, or query through an agent connected to anindexed_sources_mcpknowledge source via/openai/v1/responses. You can query multiple collections in a single request.
For more information, see Collections overview and MCP Server overview.
Performance considerations
Each search type offers different performance characteristics. Consider these points when choosing the best option for your scenario:
Text, vector, and hybrid search: These search types are optimized for speed and efficiency. They return results quickly, making them a good choice for most standard Q&A and document lookup scenarios.
Hybrid multimodal search: This search type processes both text and images, which can increase processing time compared to text-only searches. Performance depends on the size and type of your data.
Tips for choosing a search type
Use these tips to help you select the best search type for your Agents and Tools with Foundry Local deployment and scenario:
- Pick the search type that matches your data and user needs.
- Use hybrid search as the default for most scenarios.
- Text, vector, and hybrid search work in all deployments. Hybrid multimodal search requires GPU mode (
gpu_enabled=trueso image embedding collections are provisioned) and, in the developer portal, a configuredbyom.apiEndpointwith a vision-capable model. - When you use agents, the built-in MCP server provides search tools. See MCP Server overview.
Search type parameters
The following tables list and describe all parameters you can configure for each search type in Agents and Tools with Foundry Local. Not all parameters are available for every search type or deployment. The Agents and Tools with Foundry Local developer portal shows only the options that apply to your selected search type and model.
Model parameters
The following model parameters are available for text, vector, hybrid, and hybrid multimodal search types. These settings control how the language model generates answers, such as the creativity or focus of responses, rather than how documents are retrieved or ranked.
| Model parameter | Description |
|---|---|
| Temperature | Controls randomness. Lowering the temperature means that the model produces more repetitive and deterministic responses. Increasing the temperature results in more unexpected or creative responses. Try adjusting temperature or Top P but not both. |
| Top-P (nucleus sampling) | Similar to temperature, this parameter controls randomness but uses a different method. Lowering Top P narrows the model's token selection to likelier tokens. Increasing Top P lets the model choose from tokens with both high and low likelihood. Try adjusting temperature or Top P but not both. |
Search parameters
The following search parameters are available for text, vector, hybrid, and hybrid multimodal search types. While model parameters control how the language model generates answers, these search parameters let you fine-tune how Agents and Tools with Foundry Local retrieves, filters, and ranks documents from your indexed data. These parameters help you optimize the relevance and precision of the information sent to the model for each query.
These parameters only apply to the Knowledge-base chat experience.
| Search parameter | Description |
|---|---|
| Top-N documents | Refers to the number of most relevant chunks given to the language model. Choose N based on the application's needs: - For broad coverage, use a larger N, such as 10-20. - For precision-critical tasks, use a smaller N, such as 3-5. Experiment with N values to balance retrieval accuracy and downstream model performance. Larger values of N increase inference time. |
| Text strictness | Controls how strictly the RAG system filters and ranks retrieved text documents before passing them to the generation model. - Low strictness: The system considers more documents relevant, even if they're only loosely related to the query. - High strictness: The system uses only documents that closely match the query. |
| Image strictness (Hybrid multimodal search only) | Similar to text strictness but applies to retrieved image data. - Low strictness: The system might retrieve a broader range of images. - High strictness: The system retrieves only images that are closely aligned with the query. |