SEARCHABLE OBJECT STORAGE
Search objects by fields, text, vectors or paths.Search objects by fields, text, vectors or paths.
Keep the original objects and add the right way to find them. Keldra supports path, metadata, typed JSON, full-text, vector, hybrid, Git-source and tensor indexes over the data your application supplies.
Find the right object, not just its address.
An object path is useful when you already know it. An index helps answer the next question without downloading every file and searching it in application code.
Find overdue tasks. Search support articles. Match similar items. Locate a file at a Git commit or a named model tensor.
Choose the index for your data
Different questions need different indexed representations. Select the type and declared capabilities that match the workload.
Object paths and prefixes
Find objects by their stored paths and prefixes. Use a structured object namespace as a useful access path.
Useful for workspace files, folders and named collections.
Typed JSON fields
Declare the JSON fields your application filters, groups or orders by. Use the exact, range and other supported capabilities those fields need.
Useful for tasks, dates, business records and application state.
Full-text search
Search analysed words and phrases in the declared text fields. Combine text matching with the supported filters your application needs.
Useful for articles, notes, descriptions and support content.
Vector similarity
Index numerical embeddings supplied by your application and retrieve similar items. Choose the embedding model that fits your data.
Useful for semantic retrieval, recommendations and visual similarity with suitable embeddings.
Hybrid text and vector search
Combine text relevance and vector similarity when neither keywords nor embeddings alone capture the whole question.
Useful when users search by both specific wording and broader meaning.
Git commits and tree paths
Look up exact or prefix tree paths at a selected commit within a configured repository. Git HTTP hosting is available alongside this source index.
Useful for repository-backed applications; not arbitrary source-text or symbol search.
Named model tensors
Look up a tensor name within a configured model using descriptors supplied by your tooling, including the stored source reference.
Useful for locating named weights; not automatic model parsing or numerical inference.
Object paths and prefixes
Find objects by their stored paths and prefixes. Use a structured object namespace as a useful access path.
Useful for workspace files, folders and named collections.
Typed JSON fields
Declare the JSON fields your application filters, groups or orders by. Use the exact, range and other supported capabilities those fields need.
Useful for tasks, dates, business records and application state.
Full-text search
Search analysed words and phrases in the declared text fields. Combine text matching with the supported filters your application needs.
Useful for articles, notes, descriptions and support content.
Vector similarity
Index numerical embeddings supplied by your application and retrieve similar items. Choose the embedding model that fits your data.
Useful for semantic retrieval, recommendations and visual similarity with suitable embeddings.
Hybrid text and vector search
Combine text relevance and vector similarity when neither keywords nor embeddings alone capture the whole question.
Useful when users search by both specific wording and broader meaning.
Git commits and tree paths
Look up exact or prefix tree paths at a selected commit within a configured repository. Git HTTP hosting is available alongside this source index.
Useful for repository-backed applications; not arbitrary source-text or symbol search.
Named model tensors
Look up a tensor name within a configured model using descriptors supplied by your tooling, including the stored source reference.
Useful for locating named weights; not automatic model parsing or numerical inference.
Object paths and prefixes
Find objects by their stored paths and prefixes. Use a structured object namespace as a useful access path.
Useful for workspace files, folders and named collections.
Typed JSON fields
Declare the JSON fields your application filters, groups or orders by. Use the exact, range and other supported capabilities those fields need.
Useful for tasks, dates, business records and application state.
Full-text search
Search analysed words and phrases in the declared text fields. Combine text matching with the supported filters your application needs.
Useful for articles, notes, descriptions and support content.
Vector similarity
Index numerical embeddings supplied by your application and retrieve similar items. Choose the embedding model that fits your data.
Useful for semantic retrieval, recommendations and visual similarity with suitable embeddings.
Hybrid text and vector search
Combine text relevance and vector similarity when neither keywords nor embeddings alone capture the whole question.
Useful when users search by both specific wording and broader meaning.
Git commits and tree paths
Look up exact or prefix tree paths at a selected commit within a configured repository. Git HTTP hosting is available alongside this source index.
Useful for repository-backed applications; not arbitrary source-text or symbol search.
Named model tensors
Look up a tensor name within a configured model using descriptors supplied by your tooling, including the stored source reference.
Useful for locating named weights; not automatic model parsing or numerical inference.
From stored data to a useful query
Declare the index first
Choose the input fields and capabilities the query needs. Create the definition before importing supported data so incoming writes follow that indexing path.
Store the original objects
Upload through the supported object APIs. Your application supplies embeddings, model descriptors or other required inputs; indexing does not automatically generate them from arbitrary files.
Query the declared representation
Use the appropriate typed query, filters, order and page boundary. Fetch original payloads when the application needs the files, not merely to decide which records match.
Index the operation you intend to use
Exact lookup, ranges, ordering and grouping are distinct capabilities. A field indexed only for grouping is not automatically an exact-lookup field. Declare the capabilities your filters, aggregations and keyed joins require.
Keep visibility part of the query
Realm-aware indexed queries can evaluate the selected subject and relation before collecting results and aggregates. Each query input can use the access rule appropriate to that object type.
Choose the index your next query needs.
Run Keldra, upload representative data and declare the index your application actually needs. Keep the original files and test the query against your own workload.