Everything you need to integrate Candengo Vector into your application. Full API reference and quickstart guides.
Interactive FastAPI docs. Try every endpoint directly in your browser with live request and response examples.
Step-by-step guide: create a namespace, ingest your first document, and run your first semantic search in under 10 minutes.
Coming soonAPI Overview
The Candengo Vector API is a REST service with JSON request and response bodies. All endpoints require authentication.
All requests must include the header X-Service-Secret: <your_key>. Your key is available from the platform UI under Settings → API Keys. On self-hosted deployments the key is configured via the CANDENGO_SECRET environment variable.
Ingest one or more documents into a namespace. Accepts plain text, structured JSON chunks, or file references. Documents are automatically split, embedded with BGE-M3 and indexed in the vector store. Supports optional metadata fields for filtering at query time.
Semantic vector search over a namespace. Pass a natural-language query and receive ranked chunks with similarity scores and source metadata. Supports top_k, metadata filter expressions, and an optional score_threshold. Hybrid dense+sparse search available on Startup and above.
Full RAG pipeline in a single call. Retrieves the most relevant chunks via /search, then forwards them as context to your configured LLM provider (OpenAI, Anthropic, xAI). Returns the generated answer together with citations pointing back to source chunks — so every answer is auditable.
List all namespaces accessible to the authenticated key, with vector counts, document counts and creation timestamps.
Create a new namespace. Specify a name, optional description, and embedding config overrides. The namespace is immediately available for ingestion.
Permanently delete a namespace and all its vectors. This action is irreversible. Workpack-managed namespaces cannot be deleted via this endpoint — uninstall the workpack from the App Store instead.
Returns service health and version information. Does not require authentication. Suitable for load-balancer health checks and uptime monitoring.