SDKs and CLI

VectorAmp SDKs and command-line tools wrap the public REST API with typed clients, resource objects, ingestion helpers, RAG/query utilities, and terminal workflows.

Languages and packages

All clients are licensed under Apache-2.0 and install from their public package registries.

SDK	Runtime	Package / module	Registry
Python	Python 3.9+	vectoramp	PyPI
Node.js	Node.js 18+	@vectoramp/vectoramp	npm
Ruby	Ruby 3.1+	vector_amp	RubyGems
Go	Go 1.18+	github.com/vectoramp/vectoramp-go	pkg.go.dev
Rust	Rust 1.75+	vectoramp	crates.io
Java	Java 11+	com.vectoramp:vectoramp-java	Maven Central
CLI	Node.js 18+	@vectoramp/cli	npm
LangChain	Python 3.9+	langchain-vectoramp	PyPI

Common SDK model

All current SDKs share the same core ideas:

• API-key authentication via X-API-Key or VECTORAMP_API_KEY.
• https://api.vectoramp.com as the default API host, with baseUrl / base_url overrides available for dedicated gateways.
• Dataset creation defaults to the SABLE index; SDKs intentionally do not expose index-type selection.
• One-call create: create("name") (or create({ name })) works with only a name — the managed VectorAmp-Embedding-4B model, an inferred dimension (2560), cosine metric, and the SABLE index are all defaulted. hybrid enables dense + sparse; openai("small"|"large") selects an OpenAI embedding with its dimension inferred.
• Object → method calls, where returned dataset objects can search, insert, addTexts / add_texts, ask, and start ingestion directly.
• Service-style calls, where you pass dataset_id explicitly through client.datasets, client.ingestion, or equivalent services.
• The full set of typed ingestion source builders — web, s3, gcs, gdrive, jira, confluence, and file_upload — plus a generic escape hatch for future source types.
• Minimal local-file ingestion that auto-creates a file_upload source when you do not pass an existing source id or source name.
• Dataset document listing/downloading for retained originals across every SDK and the CLI.
• Numeric or string vector ids: integer ids are serialized as JSON numbers and preserved by the API.
• Persistent Intelligence sessions (create_session / createSession, append_message, list_messages, list_sessions) in every SDK.

Search rerank support

VectorAmp exposes two different search rerank knobs, and every SDK and the CLI support both:

• rerank: true (equivalently rerank: { enabled: true }, RerankConfig{Enabled: true}, or with_rerank(true)) enables the optional model reranker after initial retrieval. It is off by default; provider/model default to vectoramp / VectorAmp-Rerank-v1, and top_n, max_documents, and fail_open use server defaults unless explicitly supplied.
• rerank_depth_override tunes SABLE's vector-index candidate-depth rerank only. It is not the model reranker.

See each language page for the idiomatic spelling. For knobs an SDK has not yet modeled, the raw REST API is always available.

Common flow

create/get dataset
  -> add_texts or ingest files/sources
  -> search by text or vector
  -> ask questions over the dataset

Use the language pages for install commands and idiomatic examples. Use the CLI guide for terminal-based dataset management, ingestion, search, and interactive ask workflows.

Every client exposes persistent Intelligence sessions and message helpers alongside ask / streaming RAG.