Skip to content

components.md

Latest commit

 

History

History
117 lines (84 loc) · 3.86 KB

components.md

File metadata and controls

117 lines (84 loc) · 3.86 KB

Components Reference

Ragtime ships with ready-to-use components: an Albert API client, two frontend applications, and pluggable modules.

Albert Client SDK

Official Python SDK for the Albert API. OpenAI-compatible with features specific to French government administration.

Installation:

pip install albert-client

Usage Example:

from albert_client import AlbertClient

client = AlbertClient(
    api_key="your-api-key",
    base_url="https://albert.api.etalab.gouv.fr/v1"
)

# OpenAI-compatible
response = client.chat.completions.create(
    model="openweight-small",
    messages=[{"role": "user", "content": "Hello!"}]
)

# Hybrid search with reranking
results = client.search(
    query="energy transition",
    collections=[1, 2],
    method="hybrid"
)

Full Documentation: Albert Client SDK README

Frontend Applications

App Description Default Port
Chainlit Chat Chat interface for querying ingested documents 8000
Reflex Chat Interactive chat with modern UI 3000

Both are available during ragtime setup and are pre-configured to work with the Albert API out of the box.

Pipeline Packages

Ragtime organizes the RAG pipeline into dedicated packages, each handling a specific phase:

Package Phase Description
ingestion Document Ingestion Parse documents into text (PDF, Markdown, HTML)
storage Vector Storage Collection management — create, populate, delete, list
retrieval Search Vector search across document collections (semantic, lexical, hybrid)
reranking Reranking Re-score search results with a cross-encoder for higher precision
context Context Assembly Format retrieved chunks into LLM-ready context strings with citations
tracing Observability Log queries, context, responses, and user feedback for pipeline improvement
pipelines Orchestration Coordinate all pipeline phases into a unified interface

Storage

Manages vector store collections via the Albert API (or local SQLite, planned).

from ragtime.storage import get_provider

provider = get_provider()  # Backend from ragtime.toml
collection_id = provider.create_collection(client, "my-docs")
provider.ingest_documents(client, ["report.pdf"], collection_id)

Retrieval → Reranking → Context

Each phase is a self-contained module under ragtime.*. The pipelines module orchestrates them:

# Individual phase modules (used by pipelines internally)
from ragtime.retrieval import search_chunks
from ragtime.reranking import rerank_chunks
from ragtime.context import format_context

chunks = search_chunks(client, "energy transition", collection_ids=[1])
reranked = rerank_chunks(client, "energy transition", chunks)
context_str = format_context(reranked)

Backend Selection

Backend is configured in ragtime.toml:

[storage]
provider = "albert-collections"  # or "local-sqlite" (planned)

To switch backends:

  1. Edit ragtime.toml
  2. Restart your application

No code changes or reinstallation needed!

Comparison Table

Feature Basic (local-sqlite) Albert (albert-collections)
Extraction Local pypdf Albert API + fallback
Formats PDF only PDF, MD, HTML
Search None (context injection) Semantic + Hybrid + Reranking
Persistence None (per-session) Collections (persistent)
Network Offline Requires API access
Use Case Small docs, prototypes Production, large collections

Note: Both backends implement the same interface, making them fully interchangeable. Apps automatically work with either backend without code changes.