Installation

Lasagna runs on Python 3.10 to 3.13.

pip install lasagna-rag                    # core: ingest, embed, search, serve
python -m nltk.downloader punkt_tab        # once: nltk's sentence tokenizer data

Lasagna never downloads the tokenizer data by itself. Without it, sentences are split with a regular expression, which can move chunk boundaries. Run bundles record that as the sentence_split_without_punkt fallback.

The package is lasagna-rag

pip install lasagna installs a different project (lasagna-ai, an LLM-agent framework). It also provides a top-level lasagna import package, so installing both in one environment makes them overwrite each other.

What the first run downloads

The default profile pulls an embedding model on the first upload or search and a reranker on the first search that reranks. That is about 3.2 GB: BGE-M3 (2.3 GB) and the mxbai reranker (881 MB). The flag-embedding extra adds 2.3 GB more, because it fetches BGE-M3's whole repository, ONNX export included. Expect between 2 and 40 minutes to your first search result on release-gate runs. The spread is your connection.

To try Lasagna faster, start on the speed profile: dense vector search on a 92 MB model, no reranker and no API key.

lasagna configure --profile speed

Configuration lists the five profiles and what each one downloads.

Pin the versions the release was tested with

For a result you want to repeat later, install with the constraints file from the release. It was resolved on Python 3.12 and needs 3.12 or later, because numpy does not install on 3.10 or 3.11 with those pins.

pip install "lasagna-rag[eval]==0.1.0" -c https://raw.githubusercontent.com/dolev31/lasagna/v0.1.0/constraints/0.1.0.txt

Extras

The core install covers ingest, embedding, search and serving. Add an extra for anything else.

Extra	Adds
eval	RAGAS metrics, benchmark loaders, statistics, plots
ingest	EasyOCR for scanned documents
docling	structure-aware document conversion (the docling processor)
markitdown	Microsoft MarkItDown: PDF, Office and HTML to Markdown (the markitdown processor)
pymupdf4llm	PDF to Markdown (the pymupdf4llm processor); AGPL-3.0
vector-backends	Pinecone, Qdrant, Weaviate clients
opensearch / pgvector	those backends
flag-embedding	BGE-M3 dense+sparse+ColBERT
graphrag	entity-linked retrieval
llamaindex	wrapping LlamaIndex components as stages
storage-all	S3, GCS, Azure document storage
local	local LLM inference (HF, Ollama)
providers	LLM providers beyond OpenAI: Anthropic, Google, Mistral, Cohere, NVIDIA
ocr	Tesseract and PaddleOCR engines (Tesseract also needs the tesseract program)
web	web search fallback for corrective RAG
lifecycle	scheduled document expiry and cleanup
production	Redis, Celery, Prometheus, OpenTelemetry and PostgreSQL: what settings.lasagna_production.toml turns on
all	everything above except markitdown, pymupdf4llm and llamaindex

A few components need a package or an API key of their own: the flashrank reranker (pip install flashrank), the cohere reranker and embedder (pip install cohere, COHERE_API_KEY), the voyage embedder (pip install voyageai, VOYAGE_API_KEY) and the openai embedder (OPENAI_API_KEY). Each raises an error that names what is missing.

all is large. pip takes about two minutes to resolve it (measured with pip 26.2), and uv pip install "lasagna-rag[all]" a few seconds.

An LLM for the generation track

Retrieval comparisons need no LLM. The generation track, the judged metrics and LLM-based components (HyDE, an LLM reranker, an agentic pipeline) do.

With no LLM configured, Lasagna uses the first API key it finds in the environment, in the order of the provider table, with that provider's default model. Set MODEL_NAME to pick another model on that provider.

• The core install talks to OpenAI, Azure OpenAI and OpenAI-compatible endpoints.
• Anthropic, Google, Mistral, Cohere and NVIDIA need the providers extra.
• A local Hugging Face model needs the local extra.
• Groq, watsonx and llama.cpp need their LangChain package.

In an experiment, the answers come from RAG_EVAL_GENERATOR_MODEL, else MODEL_NAME, else the provider default. The judge is separate: RAG_EVAL_JUDGE_MODEL, else judge_model: in the sweep file, else gpt-6-luna. To write your own provider, see the providers package.

Platform notes

On Linux x86_64, pip installs the CUDA build of torch, about 3.2 GB of downloads. For CPU only, install torch from the PyTorch CPU index first:

pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install lasagna-rag

On Apple Silicon, model calls on the GPU (MPS) run one at a time, because two at once crash Metal. Concurrent sweep arms wait for each other there. To run everything on the CPU instead, set RAG_USE_GPU=false. The install weighs about 1.3 GB on macOS arm64.

import lasagna loads no torch, transformers, sentence-transformers or pymilvus. Those arrive only when something needs them, so a remote-backend client stays light.

Telemetry

Lasagna collects no usage data. ragas, the library behind the generation metrics, sends its authors a usage event (its version, the LLM provider and model name, and a random id) each time a judge is built. Importing lasagna turns that off by setting RAGAS_DO_NOT_TRACK=true, unless you have already set it. To send ragas those events, set RAGAS_DO_NOT_TRACK=false before importing lasagna. ragas still writes its random id to a file in your user data folder, but with tracking off it sends nothing.

Check the install

lasagna doctor

lasagna doctor checks the install and names what is missing. From a source checkout, python -m e2e.run measures the cold start (the e2e package is not in the wheel).

Next

Quickstart uploads a file and searches it.