Keystone knowledge base

Use the Keystone REST API to import content and retrieve grounded context from the user's knowledge bases. Never expose the API key in messages, commands logged to shared output, or saved files.

Setup

1. In Keystone, open Settings → API Integration and create or copy an API key.
2. Configure the agent environment with the public Keystone API address. The address must end with /api/v1 and be reachable from the agent runtime.

export KEYSTONE_BASE_URL="https://keystone.example.com/api/v1"
export KEYSTONE_API_KEY="sk-your-api-key"

For a local deployment used by an agent on the same computer, the usual base URL is http://localhost:8080/api/v1.

Credential check

Before making an API request, ensure both values exist. If either is missing, ask the user to configure it; do not guess or substitute a token.

if [ -z "$KEYSTONE_BASE_URL" ] || [ -z "$KEYSTONE_API_KEY" ]; then
  echo "Missing Keystone credentials. Set KEYSTONE_BASE_URL and KEYSTONE_API_KEY."
  exit 1
fi

Request helper

All JSON API requests use X-API-Key. Keep the endpoint relative to the base URL so deployments behind a reverse proxy continue to work.

keystone_api() {
  local method="$1" endpoint="$2" body="$3"
  curl --fail-with-body -sS -X "$method" "$KEYSTONE_BASE_URL/$endpoint" \
    -H "X-API-Key: $KEYSTONE_API_KEY" \
    -H "Content-Type: application/json" \
    -H "X-Request-ID: $(uuidgen 2>/dev/null || date +%s)" \
    ${body:+-d "$body"}
}

For uploads, use curl -F directly. Do not set Content-Type manually for a multipart request.

Choose the right API

User intent	Endpoint	Notes
List knowledge bases	GET /knowledge-bases	Select a KB by id or name before importing/searching.
View KB details	GET /knowledge-bases/:id	Inspect indexing and configuration.
Upload a file	POST /knowledge-bases/:id/knowledge/file	Multipart field: file; optional enable_multimodel.
Import a web page	POST /knowledge-bases/:id/knowledge/url	JSON: url, optional enable_multimodel.
Create Markdown knowledge	POST /knowledge-bases/:id/knowledge/manual	JSON: title, content, optional tag_id.
Check processing	GET /knowledge/:id	Poll parse_status after an import.
Browse KB entries	GET /knowledge-bases/:id/knowledge	Use page, page_size, optional tag_id.
Edit Markdown knowledge	PUT /knowledge/manual/:id	JSON: title, content.
Delete a knowledge entry	DELETE /knowledge/:id	Confirm destructive actions with the user first.
Search one KB	GET /knowledge-bases/:id/hybrid-search	JSON body: query_text, match_count, thresholds.
Search several KBs	POST /knowledge-search	JSON: query, knowledge_base_ids.

Common workflows

Upload a file and wait for parsing

# First find the target knowledge base and its id.
keystone_api GET "knowledge-bases"

# Upload. The response contains data.id (knowledge id).
curl --fail-with-body -sS -X POST "$KEYSTONE_BASE_URL/knowledge-bases/<kb_id>/knowledge/file" \
  -H "X-API-Key: $KEYSTONE_API_KEY" \
  -F 'file=@document.pdf' \
  -F 'enable_multimodel=true'

# Poll until data.parse_status is completed or failed.
keystone_api GET "knowledge/<knowledge_id>"

Import a URL or Markdown

keystone_api POST "knowledge-bases/<kb_id>/knowledge/url" \
  '{"url":"https://example.com/article","enable_multimodel":true}'

keystone_api POST "knowledge-bases/<kb_id>/knowledge/manual" \
  '{"title":"Meeting notes","content":"# Q1 review\n\nKey points..."}'

Retrieve knowledge

# Hybrid retrieval within one knowledge base. This GET endpoint expects a JSON body.
keystone_api GET "knowledge-bases/<kb_id>/hybrid-search" \
  '{"query_text":"deployment process","match_count":5}'

# Search across selected knowledge bases.
keystone_api POST "knowledge-search" \
  '{"query":"deployment process","knowledge_base_ids":["kb-1","kb-2"]}'

Response handling

• Successful responses put data in data (lists are usually data[]).
• Knowledge processing progresses from pending to processing, then completed or failed. Check error_message before retrying a failure.
• Hybrid-search results include content, score, knowledge_id, knowledge_title, and chunk metadata. Use them as source context and say when no relevant result was returned.
• Paginated knowledge lists return total, page, and page_size.

Safety and error handling

• Do not upload, import, overwrite, or delete anything without the user's explicit target knowledge base and confirmation for destructive actions.
• Treat 401 as invalid/missing API credentials, 403 as insufficient access, 404 as a wrong resource id, 413 as an oversized upload, and 429 as a rate limit requiring a pause before retry.
• For a failed parse, inspect error_message; retry with POST /knowledge/:id/reparse only when the user asks to retry.