Tenants

Provisions a new tenant and returns its id.#

POST/lmkit/v1/search/tenants

The server generates the tenant id and returns it. Use it to create collections, index documents, and search on behalf of this tenant.

Request body

application/json ·

Property	Type	Description
`cluster_id`	string	The search cluster to create the tenant in. If not provided, the default cluster is used.
`display_name`	string	Optional human-readable name for the tenant.
`embedding_model`	string	Optional embedding model for the tenant: one of the cluster's model catalog, by default 'embeddinggemma-300m', 'qwen3-embedding:0.6b', 'qwen3-embedding:4b', 'qwen3-embedding:8b', or 'harrier-oss:0.6b'. The model is a tenant-wide property: every collection that enables semantic search is embedded with it. Omit to provision a full-text-only tenant; set or change it later via the embedding-model endpoint.
`enable_ocr`	boolean	Whether the tenant runs OCR while extracting scanned pages and images. Defaults to true. Tenant-wide and applied to newly indexed documents; change it later via the tenant update endpoint.
`ocr_noise_rejection`	boolean
`enable_full_text_search`	boolean	Whether the tenant's collections support full-text (word-based) search. Defaults to true. Tenant-wide, so every collection shares it. A tenant must keep at least one search mode enabled.
`enable_semantic_search`	boolean	Whether the tenant's collections support semantic (meaning-based) search. Defaults to false. Tenant-wide; enabling it embeds documents with the tenant's embedding model. A tenant must keep at least one search mode enabled.
`normalization`		Optional text-normalization options (case / accent / unicode / punctuation folding), applied to every collection. Omit for the defaults (case-insensitive, accent-sensitive); change it later via the tenant update endpoint.
`rerank_model`	string	Optional reranking model for the tenant, for example 'bge-m3-reranker'. The model is a tenant-wide property used by the optional second-stage reranking pass (a search request with 'rerank' set), which re-scores candidates by direct query-document relevance to lift precision in every mode. Omit to leave the tenant without a reranker; set or change it later via the tenant update endpoint.
`query_model`	string	Optional query-understanding model for the tenant: the small local LLM used by search requests with a 'query_mode' beyond Original (contextual rewriting, multi-query expansion, HyDE). A small instruct model (1B class) keeps the added latency low. Omit to leave the tenant without a query model (such requests then degrade to the original query); set or change it later via the tenant update endpoint.

Responses

Status	Type	Description
201		Created
404		Not Found
503	application/json	Service Unavailable

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/tenants" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "display_name": "",
  "embedding_model": "string",
  "enable_ocr": true
}'

Lists the tenants in a cluster.#

GET/lmkit/v1/search/clusters/{clusterId}/tenants

Parameters

Name	In	Type	Description
`clusterId`required	path	string

Responses

Status	Type	Description
200	[]	OK
404		Not Found
503	application/json	Service Unavailable

curl -X GET "$LMKIT_ONE_URL/lmkit/v1/search/clusters/$CLUSTERID/tenants" \
  -H "Authorization: Bearer $LMKIT_API_KEY"

Gets a tenant.#

GET/lmkit/v1/search/clusters/{clusterId}/tenants/{tenantId}

Parameters

Name	In	Type	Description
`clusterId`required	path	string
`tenantId`required	path	string (uuid)

Responses

Status	Type	Description
200		OK
404		Not Found
503	application/json	Service Unavailable

curl -X GET "$LMKIT_ONE_URL/lmkit/v1/search/clusters/$CLUSTERID/tenants/$TENANTID" \
  -H "Authorization: Bearer $LMKIT_API_KEY"

Updates a tenant's configuration (embedding model, OCR, search modes, normalization).#

PUT/lmkit/v1/search/clusters/{clusterId}/tenants/{tenantId}

Updates the tenant's settings in one call. Every setting is tenant-wide and shared by all the tenant's collections. This replaces the configuration as a whole, so send the full desired state. Changes that affect indexed documents are applied in the background and search keeps working: changing the embedding model (or turning semantic on) re-embeds documents, and changing normalization re-indexes full text. An empty embedding model clears semantic capability and drops vector data. A tenant must keep at least one search mode enabled.

Parameters

Name	In	Type	Description
`clusterId`required	path	string
`tenantId`required	path	string (uuid)

Request body

application/json ·

Property	Type	Description
`embedding_model`	string	The tenant's embedding model: one of the cluster's model catalog, by default 'embeddinggemma-300m', 'qwen3-embedding:0.6b', 'qwen3-embedding:4b', 'qwen3-embedding:8b', or 'harrier-oss:0.6b'. Null or omitted leaves the current model unchanged; the model is never cleared through this call.
`enable_ocr`	boolean	Whether the tenant runs OCR while extracting scanned pages and images during indexing. Null or omitted leaves it unchanged.
`ocr_noise_rejection`	boolean
`enable_full_text_search`	boolean	Whether the tenant's collections support full-text search. Null or omitted leaves it unchanged. A tenant must keep at least one search mode enabled.
`enable_semantic_search`	boolean	Whether the tenant's collections support semantic search (requires an embedding model). Null or omitted leaves it unchanged. A tenant must keep at least one search mode enabled.
`normalization`		Text-normalization options (case / accent / unicode / punctuation folding). Null or omitted leaves them unchanged.
`rerank_model`	string	The tenant's reranking model (for example 'bge-m3-reranker'), used by the optional second-stage reranking pass. Null or omitted leaves it unchanged; send an empty string to clear it (the tenant then has no reranker).
`query_model`	string	The tenant's query-understanding model: the small local LLM used by search requests with a 'query_mode' beyond Original (contextual rewriting, multi-query expansion, HyDE). Null or omitted leaves it unchanged; send an empty string to clear it (such requests then degrade to the original query).

Responses

Status	Type	Description
200		OK
400		Bad Request
404		Not Found
503	application/json	Service Unavailable

curl -X PUT "$LMKIT_ONE_URL/lmkit/v1/search/clusters/$CLUSTERID/tenants/$TENANTID" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "embedding_model": "string",
  "enable_ocr": false,
  "ocr_noise_rejection": false,
  "enable_full_text_search": false
}'

Offboards a tenant, deleting all of its indexed data.#

DELETE/lmkit/v1/search/clusters/{clusterId}/tenants/{tenantId}

Parameters

Name	In	Type	Description
`clusterId`required	path	string
`tenantId`required	path	string (uuid)

Responses

Status	Type	Description
204		No Content
404		Not Found
503		Service Unavailable

curl -X DELETE "$LMKIT_ONE_URL/lmkit/v1/search/clusters/$CLUSTERID/tenants/$TENANTID" \
  -H "Authorization: Bearer $LMKIT_API_KEY"

Gives a large tenant its own dedicated vector index partition.#

POST/lmkit/v1/search/clusters/{clusterId}/tenants/{tenantId}/promote

Optional scaling step for high-volume tenants. By default a tenant's semantic-search vectors share a pooled index with other small tenants. This moves the tenant's vectors into a dedicated partition, so its vector (nearest-neighbor) index holds only its own data, which keeps semantic search fast and accurate as the tenant's document count grows. Apply it before bulk-indexing a large tenant. It has no effect on small tenants and is never required for correctness.

Parameters

Name	In	Type	Description
`clusterId`required	path	string
`tenantId`required	path	string (uuid)

Responses

Status	Type	Description
204		No Content
404		Not Found
503		Service Unavailable

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/clusters/$CLUSTERID/tenants/$TENANTID/promote" \
  -H "Authorization: Bearer $LMKIT_API_KEY"

Returns a tenant from its dedicated vector index partition to the shared pool.#

POST/lmkit/v1/search/clusters/{clusterId}/tenants/{tenantId}/demote

Reverses promotion: moves the tenant's semantic-search vectors back into the shared pool and drops its dedicated partition. Vectors are preserved (no re-embedding). A no-op for a tenant that was never promoted.

Parameters

Name	In	Type	Description
`clusterId`required	path	string
`tenantId`required	path	string (uuid)

Responses

Status	Type	Description
204		No Content
404		Not Found
503		Service Unavailable

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/clusters/$CLUSTERID/tenants/$TENANTID/demote" \
  -H "Authorization: Bearer $LMKIT_API_KEY"