Knowledge bases

Worked knowledge-bases samples

The endpoint list below documents every operation’s parameters; this overlay walks a knowledge base the way an AI-agent integration actually uses it: list the knowledge bases → read their documents → upload one document → handle a stale or rejected-upload error. Success envelopes are { data, meta }, error envelopes { error, meta } — see How to read a worked sample. Run samples with a sandbox (dv_test_sk_*) key; documents the pipeline cannot validate (error indexing status) never enter the agent’s retrieval set, and you can re-queue them via the retry endpoint (see Knowledge bases). Every response carries meta.request_id and meta.timestamp. Quote the request id when you report a failing upload or a drifted document status.

1. List knowledge bases

GET /api/v1/knowledge-bases returns your workspace’s knowledge-base records — id, name, description, optional config, refresh cadence — with pagination. Use this to populate an AI-agent threading UI or pick the kb to which a document will be uploaded.

curl -X GET "https://api.orbit.devotel.io/api/v1/knowledge-bases" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases",
  {
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

200

{
  "data": {
    "knowledge_bases": [
      {
        "id": "kb_8fb2d1b7c00c4ec9",
        "name": "Checkout troubleshooting",
        "description": "Returns/ refund rules / payment errors",
        "refresh_schedule": "weekly",
        "document_count": 3
      }
    ],
    "pagination": { "cursor": null, "has_more": false }
  },
  "meta": {
    "request_id": "req_kb_list",
    "timestamp": "2026-08-26T12:00:00.000Z"
  }
}

Field notes a reader will hit on first integration:

refresh_schedule is manual (default) or one of the recurring cadence values the create endpoint accepts — the non-manual values drive recurring re-scrape of URL/RSS/Notion sources.
document_count is a read cache; update it with the list- read and use the documents endpoint below for per-file truth.

2. Read a knowledge base’s documents

GET /api/v1/knowledge-bases/{id}/documents returns the document ledger for one knowledge base — the rows an operator sees on the dashboard — and reports pre-ingest status for each file (pending, indexed, error, failed). A rejected upload tells you the analysis and moderation pipeline flagged it so you can act on the offending chunk.

curl -X GET \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents",
  {
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

200

{
  "data": {
    "documents": [
      {
        "id": "doc_8fb2d1b7c00c4ec9",
        "name": "returns-policy.md",
        "type": "markdown",
        "status": "indexed",
        "version": 2,
        "created_at": "2026-08-01T12:00:00.000Z"
      }
    ]
  },
  "meta": {
    "request_id": "req_kb_documents",
    "timestamp": "2026-08-26T12:01:00.000Z"
  }
}

3. Upload one document

POST /api/v1/knowledge-bases/{id}/documents ingests a text/markdown document into the knowledge base. The body is the uploadDocumentSchema — name, type (one of the allowed text-code/image formats via vision OCR), and content up to the router’s 10 MB text-upload cap. Files larger than that need the multipart endpoint the dashboard uses; an image upload flows through OCR transcription and chains into the same retrieval set.

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "name": "returns-policy.md",
  "type": "markdown",
  "content": "# Returns — 30-day window\n\n…"
}'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents",
  {
    method: "POST",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      name: "returns-policy.md",
      type: "markdown",
      content: "# Returns — 30-day window\n\n…",
    }),
  },
);
console.log(await res.json());

200

{
  "data": {
    "id": "doc_9e8d7c6b5a4f3021",
    "status": "pending",
    "version": 1
  },
  "meta": {
    "request_id": "req_kb_upload",
    "timestamp": "2026-08-26T12:02:00.000Z"
  }
}

4. Errors

Errors follow the { error, meta } envelope. The failure every uploader hits: 422 — validation. A name past 200 characters, unsupported type, or a body over the text-upload cap:

422

{
  "error": {
    "code": "VALIDATION_ERROR",
    "message": "Document content too large (max 10MB). Split the file or upload via the multipart endpoint.",
    "status": 422
  },
  "meta": {
    "request_id": "req_kb_err",
    "timestamp": "2026-08-26T12:03:00.000Z"
  }
}

5. Worked sequences — advanced lifecycle

Beyond upload and read-back, most AI-agent deployments run the full editorial loop: register an external source connector with POST /api/v1/knowledge-bases/{id}/connectors, kick off its initial sync with POST .../sync, watch drift and staleness, moderate the AI assistant’s proposed documents, roll a published version back after reading the lineage, and delete the documents the team has retired. The samples below keep the same identifiers end-to-end; pause on scheduled syncs by patching a connector with enabled: false.

5.1 Create a connector and start its initial sync

Register a connector that pulls one knowledge base’s documents from an external source. The body names the provider (one of notion, confluence, google_drive, sharepoint, zendesk) plus an optional scope, sync_schedule (default daily), and auto_refresh_on_drift. The returned id stays the handle for every later call.

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/connectors" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "provider": "notion",
  "display_name": "Returns & refunds space",
  "scope": { "space_keys": ["RETURNS"] },
  "sync_schedule": "hourly",
  "auto_refresh_on_drift": true
}'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/connectors",
  {
    method: "POST",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      provider: "notion",
      display_name: "Returns & refunds space",
      scope: { space_keys: ["RETURNS"] },
      sync_schedule: "hourly",
      auto_refresh_on_drift: true,
    }),
  },
);
console.log(await res.json());

201

{
  "data": {
    "id": "kbc_4b5e8a1f3c9d0e2f",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "provider": "notion",
    "display_name": "Returns & refunds space",
    "scope": { "space_keys": ["RETURNS"] },
    "sync_schedule": "hourly",
    "auto_refresh_on_drift": true,
    "enabled": true,
    "created_at": "2026-08-26T12:04:00.000Z"
  },
  "meta": {
    "request_id": "req_kb_connector",
    "timestamp": "2026-08-26T12:04:01.000Z"
  }
}

Start the first pull right away instead of waiting for the schedule — the response is 202 because the sync runs asynchronously; poll GET /api/v1/knowledge-bases/{id}/connectors for the connector’s last-sync status.

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/connectors/kbc_4b5e8a1f3c9d0e2f/sync" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/connectors/kbc_4b5e8a1f3c9d0e2f/sync",
  {
    method: "POST",
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

202

{
  "data": {
    "id": "kbc_4b5e8a1f3c9d0e2f",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "status": "running",
    "last_sync_started_at": "2026-08-26T12:04:30.000Z"
  },
  "meta": {
    "request_id": "req_kb_connector_sync",
    "timestamp": "2026-08-26T12:04:30.000Z"
  }
}

5.2 Read the connector drift report and staleness

Once a sync has pulled, the connector drain worker hashes the live source and compares it against the cached content your agents ground on. Read the drift scorecard to decide whether to re-sync or to switch auto_refresh_on_drift on. Until the worker has checked a connector at least once, detection_active is false — the 0-drifted totals are not yet a passing signal.

curl -X GET \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/connector-drift-report" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/connector-drift-report",
  {
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

200

{
  "data": {
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "detection_active": true,
    "totals": {
      "connectors": 2,
      "in_sync": 1,
      "drifted": 1,
      "unknown": 0
    },
    "drifted": [
      {
        "id": "kbc_4b5e8a1f3c9d0e2f",
        "provider": "notion",
        "display_name": "Returns & refunds space",
        "enabled": true,
        "auto_refresh_on_drift": true,
        "drift_status": "drifted",
        "drift_detected_at": "2026-08-26T11:55:00.000Z",
        "last_drift_checked_at": "2026-08-26T11:55:00.000Z",
        "last_synced_at": "2026-08-26T10:00:00.000Z"
      }
    ],
    "connectors": [
      {
        "id": "kbc_4b5e8a1f3c9d0e2f",
        "provider": "notion",
        "display_name": "Returns & refunds space",
        "enabled": true,
        "auto_refresh_on_drift": true,
        "drift_status": "drifted",
        "drift_detected_at": "2026-08-26T11:55:00.000Z",
        "last_drift_checked_at": "2026-08-26T11:55:00.000Z",
        "last_synced_at": "2026-08-26T10:00:00.000Z"
      },
      {
        "id": "kbc_7e2a9d1c4f8b5e6a",
        "provider": "confluence",
        "display_name": "Checkout help",
        "enabled": true,
        "auto_refresh_on_drift": false,
        "drift_status": "in_sync",
        "drift_detected_at": null,
        "last_drift_checked_at": "2026-08-26T11:50:00.000Z",
        "last_synced_at": "2026-08-26T11:00:00.000Z"
      }
    ]
  },
  "meta": {
    "request_id": "req_kb_drift",
    "timestamp": "2026-08-26T12:05:00.000Z"
  }
}

Staleness is the document-side companion: it buckets each document into fresh, stale, or critical against your org’s threshold (90 days by default; override per call with ?default_threshold_days= or per document via its metadata) so an operator’s review queue catches the files an agent still quotes as current.

curl -X GET \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/staleness" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/staleness",
  {
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

200

{
  "data": {
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "totals": {
      "fresh": 2,
      "stale": 1,
      "critical": 0,
      "documents_scanned": 3,
      "never_verified": 1
    },
    "attention_required": [
      {
        "id": "doc_8fb2d1b7c00c4ec9",
        "name": "returns-policy.md",
        "status": "stale",
        "last_verified_at": "2026-05-15T09:00:00.000Z"
      }
    ],
    "default_threshold_days": 90,
    "scanned_at": "2026-08-26T12:05:30.000Z"
  },
  "meta": {
    "request_id": "req_kb_staleness",
    "timestamp": "2026-08-26T12:05:30.000Z"
  }
}

Both endpoints are read-only and deliberately avoid audit-log writes, so polling them from a dashboard or a scheduled job stays cheap.

5.3 Approve or reject the assistant’s proposed document

This moderation path runs in addition to the plain upload flow when your knowledge base enables require_publish_approval, or when an AI assistant (for example, a gap-cluster draft) proposes a document — either way the file arrives with status: pending and never enters retrieval until a reviewer acts. Approving flips the document to approved and starts the deferred vector indexing; rejecting flips it to rejected and its chunks are never indexed. Optional notes are stored on the document so the submitter gets actionable feedback. Approve:

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_3c9d0e2f4b5e8a1f/approve" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "notes": "Reviewed against the current returns policy." }'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_3c9d0e2f4b5e8a1f/approve",
  {
    method: "POST",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      notes: "Reviewed against the current returns policy.",
    }),
  },
);
console.log(await res.json());

200

{
  "data": {
    "id": "doc_3c9d0e2f4b5e8a1f",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "name": "Returns policy draft",
    "status": "approved"
  },
  "meta": {
    "request_id": "req_kb_approve",
    "timestamp": "2026-08-26T12:06:00.000Z"
  }
}

Reject the same way, or fix and re-upload before approving — a document that was never indexed stays out of agent retrieval either way:

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_3c9d0e2f4b5e8a1f/reject" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "notes": "Superseded by the 2026-08 policy revision." }'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_3c9d0e2f4b5e8a1f/reject",
  {
    method: "POST",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      notes: "Superseded by the 2026-08 policy revision.",
    }),
  },
);
console.log(await res.json());

200

{
  "data": {
    "id": "doc_3c9d0e2f4b5e8a1f",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "name": "Returns policy draft",
    "status": "rejected"
  },
  "meta": {
    "request_id": "req_kb_reject",
    "timestamp": "2026-08-26T12:06:10.000Z"
  }
}

Only documents pending review accept either decision — moderation status approved (or missing) returns 409 CONFLICT, and a publish-scan rejected document must be fixed and re-uploaded before it can be approved. Both actions are audit-logged.

5.4 Roll back a published version

Read the version lineage first, then the diff that tells you what the head changed, and roll back the published version if the retrieval set needs a previous copy. GET .../versions lists the immutable lineage — current head plus retained snapshots — without the heavy content blobs.

curl -X GET \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/versions" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/versions",
  {
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

200

{
  "data": {
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "document_id": "doc_8fb2d1b7c00c4ec9",
    "current": {
      "version": 2,
      "name": "returns-policy.md",
      "type": "markdown",
      "char_count": 1820,
      "is_current": true
    },
    "history": [
      {
        "version": 1,
        "name": "returns-policy.md",
        "type": "markdown",
        "char_count": 1740,
        "is_current": false
      }
    ]
  },
  "meta": {
    "request_id": "req_kb_versions",
    "timestamp": "2026-08-26T12:07:00.000Z"
  }
}

Read the diff for the two versions you care about. A version outside the retained lineage returns 404; a version whose content was too large to retain returns 422 — rollback works only against retained snapshots.

curl -X GET \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/versions/diff?from=1&to=2" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/versions/diff?from=1&to=2",
  {
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(await res.json());

200

{
  "data": {
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "document_id": "doc_8fb2d1b7c00c4ec9",
    "from_version": 1,
    "to_version": 2,
    "added": 4,
    "removed": 2,
    "unchanged": 61,
    "truncated": false,
    "hunks": [
      "- Refunds take up to 10 business days.",
      "+ Returns are processed within 5–7 business days of approval.",
      "+ Enterprise contracts follow the contractual SLA instead."
    ]
  },
  "meta": {
    "request_id": "req_kb_diff",
    "timestamp": "2026-08-26T12:07:15.000Z"
  }
}

Then roll the document back. Non-destructive: the current head moves into the retained history first, so the operation is fully reversible.

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/rollback" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "version": 1 }'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/rollback",
  {
    method: "POST",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({ version: 1 }),
  },
);
console.log(await res.json());

202

{
  "data": {
    "id": "doc_8fb2d1b7c00c4ec9",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "name": "returns-policy.md",
    "status": "processing",
    "version": 3
  },
  "meta": {
    "request_id": "req_kb_rollback",
    "timestamp": "2026-08-26T12:07:30.000Z"
  }
}

5.5 Search the knowledge base

The POST .../search hook is the same retrieval an attached AI agent’s search_knowledge tool runs, and the response shape exists so you can preview what an agent would ground on before a customer asks. The body is query (required, ≤ 1000 chars) plus optional limit (default 5, max 20); when a per-base search config is active, recency, source weighting, and category filters apply before the list is truncated. An empty results array means no chunk cleared the raw similarity floor — check whether the document you expect is uploaded, approved, and indexed.

curl -X POST \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/search" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "query": "How long do annual-plan returns take?",
  "limit": 3
}'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/search",
  {
    method: "POST",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      query: "How long do annual-plan returns take?",
      limit: 3,
    }),
  },
);
console.log(await res.json());

200

{
  "data": {
    "query": "How long do annual-plan returns take?",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "results": [
      {
        "content": "Annual-plan returns are processed within 5–7 business days of approval.",
        "score": 0.87,
        "documentId": "doc_8fb2d1b7c00c4ec9",
        "document_name": "returns-policy.md",
        "chunkIndex": 2
      },
      {
        "content": "Pro-rated refunds apply to annual subscriptions cancelled mid-term.",
        "score": 0.81,
        "documentId": "doc_8fb2d1b7c00c4ec9",
        "document_name": "returns-policy.md",
        "chunkIndex": 4
      }
    ]
  },
  "meta": {
    "request_id": "req_kb_search",
    "timestamp": "2026-08-26T12:08:00.000Z"
  }
}

5.6 Delete a document and re-approve publication

A document stuck in moderation, or one the team retires, goes away in one call; anything that carries a public help-center article keeps its status through the same record — flip is_public on the publication endpoint once status is ready. Deleting returns 204 with no body.

curl -X DELETE \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_3c9d0e2f4b5e8a1f" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY"

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_3c9d0e2f4b5e8a1f",
  {
    method: "DELETE",
    headers: { "X-API-Key": process.env.ORBIT_API_KEY! },
  },
);
console.log(res.status); // 204

204

(null body — the document and its chunks are removed from retrieval)

Then re-publish the surviving document the public help center should keep. The body is { is_public: boolean, slug?, summary? }; flipping public requires the document’s ingestion be done (status: ready), and the slug is unique across every published article in your workspace — a slug collision returns 409 CONFLICT.

curl -X PATCH \
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/publication" \
  -H "X-API-Key: dv_test_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "is_public": true,
  "slug": "returns-policy",
  "summary": "How to return an order and when refunds land."
}'

const res = await fetch(
  "https://api.orbit.devotel.io/api/v1/knowledge-bases/kb_8fb2d1b7c00c4ec9/documents/doc_8fb2d1b7c00c4ec9/publication",
  {
    method: "PATCH",
    headers: {
      "X-API-Key": process.env.ORBIT_API_KEY!,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      is_public: true,
      slug: "returns-policy",
      summary: "How to return an order and when refunds land.",
    }),
  },
);
console.log(await res.json());

200

{
  "data": {
    "id": "doc_8fb2d1b7c00c4ec9",
    "knowledge_base_id": "kb_8fb2d1b7c00c4ec9",
    "name": "returns-policy.md",
    "status": "ready",
    "is_public": true,
    "slug": "returns-policy",
    "summary": "How to return an order and when refunds land."
  },
  "meta": {
    "request_id": "req_kb_publication",
    "timestamp": "2026-08-26T12:09:00.000Z"
  }
}

A document deleted after approval had its chunks removed from the vector index — a re-upload restarts the moderation queue from pending.

API Reference

All Endpoints

Worked knowledge-bases samples

1. List knowledge bases

2. Read a knowledge base’s documents

3. Upload one document

4. Errors

5. Worked sequences — advanced lifecycle

5.1 Create a connector and start its initial sync

5.2 Read the connector drift report and staleness

5.3 Approve or reject the assistant’s proposed document

5.4 Roll back a published version

5.5 Search the knowledge base

5.6 Delete a document and re-approve publication

​Worked knowledge-bases samples

​1. List knowledge bases

​2. Read a knowledge base’s documents

​3. Upload one document

​4. Errors

​5. Worked sequences — advanced lifecycle

​5.1 Create a connector and start its initial sync

​5.2 Read the connector drift report and staleness

​5.3 Approve or reject the assistant’s proposed document

​5.4 Roll back a published version

​5.5 Search the knowledge base

​5.6 Delete a document and re-approve publication

Worked knowledge-bases samples

1. List knowledge bases

2. Read a knowledge base’s documents

3. Upload one document

4. Errors

5. Worked sequences — advanced lifecycle

5.1 Create a connector and start its initial sync

5.2 Read the connector drift report and staleness

5.3 Approve or reject the assistant’s proposed document

5.4 Roll back a published version

5.5 Search the knowledge base

5.6 Delete a document and re-approve publication