Grounded LLM

What's in v0.1.0

• Hybrid retrieval — BM25 sparse + dense embeddings + RRF (RAG_RETRIEVAL_MODE=hybrid)
• Vector backends — Chroma, Qdrant, pgvector (VECTOR_STORE)
• Quality gate — 99 retrieval eval cases + adversarial suite in CI
• Platform — Go orchestration + Python retrieval (Flask), multi-tenant /api/v1, OIDC/RBAC, Helm/Terraform
• Template packs — HR, IT Support, Legal FAQ
• Ecosystem — agent runtime is a separate project, not in the standard core

What makes it a standard

• Grounded Spec v1 — normative API contract
• RFC-0001 — «Grounded-compatible» definition
• Conformance suite — run against any deployment
• Public benchmark — 99 retrieval + adversarial eval

Quick start

git clone https://github.com/kantik001/grounded-llm.git
cd grounded-llm
cp .env.example .env
docker compose up -d --build

pip install -r conformance/requirements.txt
python -m conformance spec

Full guide: Deploy · SDK

Tagged images on GHCR: ghcr.io/kantik001/grounded-llm-server:0.1.0 · -python · -webapp

Check compatibility

# Offline (OpenAPI)
python -m conformance spec

# Live server
python -m conformance check --url http://localhost:8080

# JSON output for CI
python -m conformance spec --json

Official template packs

• hr · domain default · eval default_en
• it_support · domain it_support · eval it_support
• legal_faq · domain legal_faq · eval legal_faq

All packs · registry: packs/registry.yaml

Strategy & roadmap

Standard strategy · Ecosystem · Roadmap · RFC process