Setting the file. One moment.
Subchapter 3.1
references/data-formats.mdMarkdown2 KBView on GitHub
Used with upsert_records() (Python SDK) or upsert-records (MCP). Records are automatically embedded using the index’s configured model.
JSON
[
{
"_id": "rec1",
"chunk_text": "Your text content here.",
"category": "example"
},
{
"_id": "rec2",
"chunk_text": "Another piece of text.",
"category": "example"
}
]_id — unique record identifier (required)fieldMap (e.g. chunk_text if fieldMap: {text: "chunk_text"})metadata key — put them directly on the recordUsed with upsert() (Python SDK) or pc index vector upsert (CLI).
JSON (with vectors array)
{
"vectors": [
{
"id": "vec1",
"values": [0.1, 0.2, 0.3],
"metadata": { "genre": "comedy", "year": 2021 }
},
{
"id": "vec2",
"values": [0.4, 0.5, 0.6],
"metadata": { "genre": "drama", "year": 2019 }
}
]
}JSONL (one vector per line)
{"id": "vec1", "values": [0.1, 0.2, 0.3], "metadata": {"genre": "comedy"}}
{"id": "vec2", "values": [0.4, 0.5, 0.6], "metadata": {"genre": "drama"}}id — unique vector identifier (required)values — dense vector as float array, length must match index dimension (required)metadata — arbitrary key-value pairs for filtering (optional)Used for keyword or hybrid search with sparse indexes.
{
"id": "vec1",
"values": [0.1, 0.2, 0.3],
"sparse_values": {
"indices": [10, 45, 316],
"values": [0.5, 0.3, 0.8]
},
"metadata": { "genre": "comedy" }
}sparse_values.indices — non-zero dimension indicessparse_values.values — corresponding float values, same length as indices