RAGhelm

Private RAG and agent release-control infrastructure with readiness scorecards, RAGRunManifests, deterministic gates, evaluation, and observability.

RAGhelm helps teams answer: can this private RAG or agent release ship, and why or why not?

Architecture posture: Pinecone-first reference runtime; backend-neutral release evidence layer.

Start here

Need	Page
Install and configure dependencies	Installation
Run the first local eval	Quick Start
Understand runtime components and evidence flow	System Overview
Understand technology choices and implementation status	Technology Choices
Review accepted decisions by priority and scope	Architecture
Understand repository layout	Repository Map
Use CLI commands	CLI Commands
Evaluate quality and regressions	Evaluation Overview

Core product objects

Object	Purpose	Implementation status
ReadinessScorecard	Human release decision surface: ship, needs_review, or block	Partial local demo artifact
RAGRunManifest	Machine/audit provenance artifact for what produced the evidence	Partial local demo artifact
Release policy	Deterministic gate rules for scorecard decisions	Target architecture
QueryPipeline	Canonical RAGhelm-owned reference runtime path	Target architecture
TargetAdapter	External target evidence boundary	Target architecture

Current posture

RAGhelm includes a working Python eval/retrieval/reference runtime surface and a local dashboard demo. The accepted architecture centers on private release control: scorecards, manifests, policies, target adapters, security boundaries, and BYOC-ready deployment. The ADR index distinguishes accepted decisions from implementation status so readers can separate implemented behavior from target architecture.

Local commands

uv sync --extra dev
uv run python -m raghelm eval --suite quick
uv run mkdocs build --strict
cd packages/dashboard && pnpm install && pnpm dev