from ragrails import RagRails
rag = RagRails()
Ingestion
| Method | Description |
|---|
setup_url(browser="chromium") | Install Playwright browser for URL ingestion |
scrape(url, *, mode, max_depth, max_pages, verbose, frontmatter, output_format, output_dest, output_dir, dlq) | Scrape URLs into document dicts |
parse(files, folder, *, frontmatter, output_format, output_dest, output_dir) | Parse local documents into document dicts |
fetch(url, *, title, description, method, headers, params, body, pagination, max_pages, timeout, apis, frontmatter, output_format, output_dest, output_dir) | Fetch REST API responses |
Chunking
| Method | Description |
|---|
chunk(markdown, *, title, source, chunk_size, chunk_overlap, min_chunk_length) | Split documents into chunk dicts |
Embedding
| Method | Description |
|---|
embedder(provider, model, input_type, options) | Create an embedder object |
embed(chunks, embedder, *, batch_size) | Generate embedding vectors for chunks |
Storing
| Method | Description |
|---|
store(embedded_chunks, *, vector_db, collection, url, batch_size, ensure_collection, options) | Upsert embedded chunks |
edit(chunks, embedder, *, vector_db, collection, url, batch_size, options) | Re-embed and replace chunks by ID |
delete(ids, *, vector_db, collection, url, options) | Delete chunks by ID |
Retrieval
| Method | Description |
|---|
reranker(provider, model, options) | Create a reranker object |
retrieve(query, *, embedder, vector_db, collection, url, top_k, use_query_rewrite, rewrite_llm, rewrite_context, session_context, use_rerank, reranker, rerank_top_k, options) | Retrieve relevant chunks |
Chat
| Method | Description |
|---|
llm(provider, model, *, max_tokens) | Create an LLM object |
chat(query, *, llm, embedder, vector_db, collection, url, options, reranker, history, history_compaction, query_rewrite, intent_routing, retrieval_quality, persona, retrieval_config) | Run a stateless RAG chat turn |
Pipeline
| Method | Description |
|---|
ingest(*, docs, urls, api, markdown, ingestion, chunking, embedding, storage) | Run ingestion → chunk → embed → store |
query(query, *, embedding, retrieval) | Run embed query → retrieve |
Config objects
| Class | Used with | Description |
|---|
DLQ(path, items) | scrape(dlq=...) | Dead-letter queue for retryable scrape failures |
QueryRewriteConfig(enabled, session_context, llm) | chat(query_rewrite=...) | Query rewriting config |
HistoryCompactionConfig(enabled, history_limit, keep_recent) | chat(history_compaction=...) | History summarisation config |
IntentRoutingConfig(enabled) | chat(intent_routing=...) | Intent routing config |
ChatRetrievalQualityConfig(min_retrieval_score, min_rerank_score, low_confidence_mode, max_context_chunks) | chat(retrieval_quality=...) | Retrieval quality thresholds |
Built-in providers
| Type | Providers |
|---|
| Embedding | voyage (voyage-3, voyage-3-lite, voyage-3-large) |
| Reranking | voyage (rerank-2-lite, rerank-2), bm25 (local) |
| LLM | openai (gpt-5.5, gpt-4.1, gpt-4o, gpt-4o-mini, …), anthropic (claude-opus-4-8, claude-sonnet-4-6, claude-haiku-4-5, …), google (gemini-3-pro, gemini-2.5-pro, gemini-2.5-flash, …) |
| Vector DB | qdrant, pinecone, weaviate |
Google Gemini models don’t support tool calling yet.
Extending providers
Register custom embedding and reranking providers at runtime:
from ragrails.models.embedder.registry import register_embedder
from ragrails.models.reranker.registry import register_reranker
register_embedder("my-provider", my_factory, default_model="my-model", models=("my-model",))
register_reranker("my-reranker", my_factory, default_model="my-model", models=("my-model",))
rag.embedder(provider="my-provider", ...).
→ Result types reference
