Quickstart: 30 Seconds to Production-Ready AI Infrastructure

TL;DR: Install with pip install fabra-ai. Define features or context with Python decorators. Run fabra serve. Full lineage and replay included.

Choose Your Track

🔧 ML Engineer Track "I need to serve features without Kubernetes" You're building ML models and need: Real-time feature serving Point-in-time correctness for training Context Records that prove what data your model saw Start here: Feature Store Quickstart

🤖 AI Engineer Track "I need to prove what my AI knew" You're building LLM apps and need: Vector search and retrieval Token budget management Immutable Context Records for compliance and debugging Start here: Context Store Quickstart

---

Feature Store in 30 Seconds

For ML Engineers — Serve features without Kubernetes, Spark, or YAML.

Fastest Path: `fabra demo`

pip install fabra-ai && fabra demo --mode features

That's it. Server starts, makes a test request, shows you the result. No Docker. No config files. No API keys.

Build Your Own Features

pip install fabra-ai

# features.py
from fabra.core import FeatureStore, entity, feature
from datetime import timedelta

store = FeatureStore()

@entity(store)
class User:
    user_id: str

@feature(entity=User, refresh=timedelta(hours=1))
def purchase_count(user_id: str) -> int:
    return db.query("SELECT COUNT(*) FROM purchases WHERE user_id = ?", user_id)

@feature(entity=User, refresh=timedelta(days=1))
def user_tier(user_id: str) -> str:
    return "premium" if is_premium(user_id) else "free"

fabra serve features.py
# Server running on http://localhost:8000

curl localhost:8000/v1/features/purchase_count?entity_id=user123
# {"value": 47, "freshness_ms": 0, "served_from": "online"}

Done. No Docker. No Kubernetes. No YAML.

What you get:

• DuckDB offline store (embedded, no setup)
• In-memory online store (instant reads)
• Point-in-time correctness for training data
• Same code works in production with Postgres + Redis

Feature Store Deep Dive → | Compare vs Feast →

---

Context Store in 60 Seconds

For AI Engineers — Build RAG with audit trails and compliance.

Fastest Path (no API keys): `fabra demo`

pip install fabra-ai && fabra demo

This prints a context_id (your receipt) and the next commands to prove value: show, diff, and verify.

Build Your Own Context Assembly

pip install fabra-ai

# chatbot.py (no API keys required)
from datetime import timedelta

from fabra.core import FeatureStore, entity, feature
from fabra.retrieval import retriever
from fabra.context import context, ContextItem

store = FeatureStore()

@entity(store)
class User:
    user_id: str

@feature(entity=User, refresh=timedelta(hours=1))
def user_tier(user_id: str) -> str:
    return "premium" if hash(user_id) % 2 == 0 else "free"

@retriever(name="knowledge_base", cache_ttl=timedelta(minutes=5))
async def search_docs(query: str, top_k: int = 3) -> list[str]:
    return [
        "Fabra creates verifiable Context Records.",
        "Use @context for prompt assembly with token budgets.",
        "Use @retriever for retrieval without YAML.",
    ][:top_k]

# Assemble context with token budget
@context(store, max_tokens=4000)
async def chat_context(user_id: str, query: str) -> list[ContextItem]:
    tier = await store.get_feature("user_tier", user_id)
    docs = await search_docs(query)
    return [
        ContextItem(content="You are helpful.", priority=0, required=True),
        ContextItem(content=f"User tier: {tier}", priority=1, required=True),
        ContextItem(content="Docs:\n" + "\n".join(docs), priority=2),
    ]

fabra serve chatbot.py

What you get:

• Vector search with pgvector
• Automatic token budgeting
• Immutable Context Records with cryptographic integrity
• Full lineage tracking (what data was used)
• Context replay and verification (fabra context verify)

Context Store Deep Dive → | RAG Audit Trail →

---

Production Stack Locally

Want Postgres + Redis locally?

fabra setup
# Generates docker-compose.yml with pgvector and Redis

docker-compose up -d

Then:

FABRA_ENV=production fabra serve features.py

FAQ

Q: How do I run a feature store locally without Docker? A: Fabra uses DuckDB (embedded) and in-memory cache for local dev. Install with pip install fabra-ai, define features in Python, run fabra serve. Zero infrastructure required.

Q: What's the simplest context infrastructure for small ML teams? A: Fabra targets "Tier 2" companies (Series B-D, 10-500 engineers) who need real-time ML and LLM features but can't afford Kubernetes ops. We own the write path — ingest, index, track freshness, and serve — giving you lineage and replay that read-only frameworks can't provide.

Q: How do I migrate from Feast to something simpler? A: Fabra eliminates YAML configuration. Define features in Python with @feature decorator, same data access patterns but no infrastructure tax.

Next Steps

• Compare vs Feast
• Deploy to Production
• Context Store - RAG infrastructure for LLMs
• RAG Chatbot Tutorial - Full example