Documents

Adds or updates a document in a collection so it can be searched.#

Indexes a document and makes its content searchable. Provide the document and the ids that identify it; how the collection is searched is decided by the collection's settings. Indexing an id that already exists replaces it in place; supplying the same content_hash that is already indexed is a no-op. The response 'outcome' field reports Created, Replaced, or Unchanged. Optionally set ttl_seconds to have the document expire automatically. If indexing takes longer than the configured timeout, returns 202 Accepted with a job_id. Poll GET /lmkit/v1/jobs/ for status and results. If the supplied input cannot be read as a supported file type, returns 415 Unsupported Media Type; other invalid or unreadable input returns 400 Bad Request.

Request body

application/json · IndexDocumentRequest

Responses

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/documents" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "tenant_id": "string",
  "collection_id": "string",
  "document_id": "string",
  "input": "",
  "input_format": "PlainText",
  "name": "string"
}'

Removes a document from a collection.#

Removes the document from the collection so it no longer appears in search results. Returns 404 Not Found if the document or its collection does not exist.

Request body

application/json · DeleteDocumentRequest

Responses

curl -X DELETE "$LMKIT_ONE_URL/lmkit/v1/search/documents" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "tenant_id": "string",
  "collection_id": "string",
  "document_id": "string"
}'

Removes every document from a collection (asynchronous job).#

Deletes all documents indexed in the collection (full-text, vector, and registry rows). A large collection holds hundreds of thousands of rows, so the delete ALWAYS runs as a background job that outlives the HTTP request: this returns 202 Accepted with a job_id immediately, the deletion runs in bounded committed batches that cannot be cancelled by a client disconnect or proxy timeout, and GET /lmkit/v1/jobs/ reports progress (progress_current = documents deleted so far, progress_total = the collection's document count) and finally result . Returns 404 Not Found when the collection does not exist.

Request body

application/json · PurgeCollectionRequest

Responses

curl -X DELETE "$LMKIT_ONE_URL/lmkit/v1/search/documents/all" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "tenant_id": "string",
  "collection_id": "string"
}'

Lists the document ids a collection holds, for reconciliation.#

Returns the public document ids currently indexed in the collection, ordered, one page at a time via 'offset' and 'limit' (a page shorter than 'limit' is the last). Lets a caller diff the server's contents against its own record of what it indexed, then delete documents the server still holds that the caller no longer tracks (orphans). Returns an empty list when the collection does not exist.

Request body

application/json · ListDocumentsRequest

Responses

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/documents/list" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "tenant_id": "string",
  "collection_id": "string",
  "offset": 0,
  "limit": 200
}'

Fetches a single document by its id, with everything indexed for it.#

Returns one document addressed by its id: its registry metadata (name, source, content hash, custom metadata, page count, and timestamps) and, by default, every page's Markdown and layout. Because a document can be large, the response can be trimmed: set 'include_pages' to false to get metadata only; set 'include_markdown' or 'include_page_layout' to false to drop those parts of each page (the omitted column is not even read from storage); and pass 'page_range' (a print-style string such as "1-3,7,10-12") to return only selected pages. 'page_count' is always the document's total page count, so a caller can page through a large document. Returns 404 Not Found when the document or its collection does not exist.

Request body

application/json · GetDocumentRequest

Responses

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/documents/get" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "tenant_id": "string",
  "collection_id": "string",
  "document_id": "string",
  "include_pages": true,
  "include_markdown": true,
  "include_page_layout": true
}'

Lists the documents most similar to a given document.#

Returns the documents whose content most resembles the given document, most similar first - the document itself is never returned. The comparison uses the document vectors maintained for semantic search (no text and no inference is involved), so it requires semantic search enabled for the tenant, otherwise 400 Bad Request. Because embedding is deferred by default, a freshly indexed document may not be comparable yet: the response 'state' reports 'embedding_pending' (transient - poll again shortly) or 'no_content' (the document has nothing to compare - permanent), with empty results; results are meaningful only when it is 'ready'. Scores are cosine similarity in 0..1: near-duplicates (the same document re-scanned, re-exported, or lightly edited) typically score above 0.9. Pass 'search_collection_ids' to look across several of the tenant's collections in one call (they always share one embedding model); by default only the document's own collection is searched. 'filters' / 'filter' narrow the candidates by custom metadata with the same algebra as the search endpoint. Returns 404 Not Found when the document or its collection does not exist.

Request body

application/json · SimilarDocumentsRequest

Responses

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/search/documents/similar" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "cluster_id": "string",
  "tenant_id": "string",
  "collection_id": "string",
  "document_id": "string",
  "search_collection_ids": [
    "string"
  ],
  "top_k": 20,
  "min_score": 0
}'