TypeScript SDK: Knowledge Management

Insert text

For short content that fits in a JSON body (< ~1MB):

const result = await client.insert({
  content: "ACME Corp reported $50M revenue in Q4 2025, up 23% YoY.",
  title: "ACME Q4 Earnings",
  tags: ["finance", "acme"],
  metadata: { source: "internal-slides" },
  wait: true,  // poll job to completion before returning (default)
});

result.job_id;           // "job_abc123"
result.facts_extracted;  // 4

InsertOptions

Field	Type	Default	Description
`content`	`string`	required	The text to ingest.
`title`	`string`	—	Human-readable title. Shown in source citations.
`tags`	`string[]`	—	Tags for later filtering (future).
`metadata`	`Record<string, unknown>`	—	Free-form metadata stored with the document.
`wait`	`boolean`	`true`	If true, block until the extraction job completes.

Upload a file

PDFs, CSVs, markdown, plain text. Images are extracted from PDFs automatically.

Node
Browser / Blob
Bun

import { readFile } from "node:fs/promises";

const buf = await readFile("./earnings.pdf");
const upload = await client.uploadFile({
  file: buf,
  filename: "earnings.pdf",
  saveToMemory: true,
});

upload.upload_id;  // "upload_xyz"
upload.status;     // "completed" (because wait defaults to true)

const input = document.querySelector<HTMLInputElement>("#file")!;
const blob = input.files![0];
if (blob) {
  await client.uploadFile({ file: blob, filename: blob.name });
}

const file = Bun.file("./earnings.pdf");
await client.uploadFile({
  file: await file.arrayBuffer() as unknown as Uint8Array,
  filename: "earnings.pdf",
});

Supported file types

PDF (text + tables + images)
CSV
Markdown
Plain text (.txt)
DOCX

Max file size depends on your plan: free tier is 10MB. See user limits at runtime.

Poll async jobs manually

If you set wait: false and want to poll yourself:

const { job_id } = await client.insert({
  content: "...",
  wait: false,
});

while (true) {
  const status = await client.getJobStatus(job_id!);
  if (status.status === "completed") break;
  if (status.status === "failed") throw new Error(status.error_details ?? "ingest failed");
  await new Promise((r) => setTimeout(r, 2000));
}

Or use the built-in waitForJob:

await client.waitForJob(job_id!, { pollIntervalMs: 1000, timeoutMs: 60_000 });

JobStatus values

`status`	Meaning
`pending`	Job accepted, not started.
`chunking`	Splitting content into chunks.
`extracting`	LLM extracting facts + entities.
`storing`	Writing to Neptune + OpenSearch.
`completed`	Done. Content is now queryable.
`failed`	See `error_details`.

Upload polling

Uploads have their own status endpoint because the pipeline goes: upload → file processor → fact extraction → ready. wait: true handles polling automatically; to do it manually:

await client.waitForUpload(upload.upload_id);

Get Started

MCP Server

CLI

Python SDK

TypeScript SDK

Concepts

Guides

TypeScript SDK: Knowledge Management

Insert text

InsertOptions

Upload a file

Supported file types

Poll async jobs manually

JobStatus values

Upload polling

​Insert text

​InsertOptions

​Upload a file

​Supported file types

​Poll async jobs manually

​JobStatus values

​Upload polling

Insert text

InsertOptions

Upload a file

Supported file types

Poll async jobs manually

JobStatus values

Upload polling