Frequently Asked Questions

General

Q: Is Fabra production-ready?

A: Yes. Fabra runs on "boring technology" (Postgres + Redis) that powers 90% of the internet. It is designed for high-throughput, low-latency serving in production environments.

Q: Do I need Kubernetes?

A: No. You can deploy Fabra on Heroku, Railway, AWS ECS, or even a single EC2 instance using Docker Compose. If you want to use Kubernetes, you can, but it is not a requirement.

Q: How does Fabra compare to Feast?

A: Fabra is context infrastructure that owns the write path. We provide the same core guarantees (Point-in-Time Correctness, Async I/O) but without the complexity of Kubernetes, Spark, or Docker registries — plus lineage, replay, and auditability that Feast doesn't have. See Fabra vs Feast for a detailed comparison.

Context Store (New in v1.2.0)

Q: What is the Context Store?

A: The Context Store is context infrastructure that owns the write path for LLM applications. We ingest, index, track freshness, and serve — not just query. This enables lineage, replay, and auditability that read-only frameworks cannot provide. It provides vector search (pgvector), document indexing, and intelligent context assembly with token budgets.

Q: Do I need the Context Store for traditional ML?

A: No. The Context Store is optional and designed for LLM/RAG applications. If you're building traditional ML models (fraud detection, recommendations), the Feature Store alone is sufficient.

Q: How does Fabra compare to LangChain?

A: LangChain is a read-only framework — it queries external stores but doesn't own your data. Fabra is infrastructure that owns the write path. We ingest, index, track freshness, and serve. This means we can provide lineage, replay, and auditability that LangChain cannot. You can use Fabra as the context layer in a LangChain application (we handle storage/serving, they handle orchestration).

Q: What embedding providers are supported?

A: OpenAI (text-embedding-3-small, ada-002) and Cohere (embed-english-v3.0). Set OPENAI_API_KEY or COHERE_API_KEY environment variable.

Q: How does token budgeting work?

A: Use @context(store, max_tokens=4000) to set a budget. Each ContextItem has a priority (0=highest). Lower-priority items are truncated first when over budget. Items with required=True raise an error if they can't fit.

Technical

Q: Can Fabra handle real-time features?

A: Yes.

1. Cached Features: Served from Redis in <5ms.
2. Computed Features: Python functions executed on-the-fly (e.g., Haversine distance).
3. Streaming: You can materialize data from Kafka/Flink into Postgres, and Fabra will serve it via the SQL path.

Q: What if I outgrow a single Postgres instance?

A: You can switch your Offline Store to Snowflake, BigQuery, or Redshift just by changing the connection string. Your feature definitions and Online Store (Redis) remain exactly the same.

Q: Why not just write to Redis directly?

A: You can, but you lose:

1. Point-in-Time Correctness: Redis only knows "now". It cannot generate historical training data without leakage.
2. Schema Evolution: Changing a feature definition in Fabra is a code change. In raw Redis, it's a migration nightmare.
3. Observability: Fabra provides built-in metrics for hit rates, latency, and staleness.

Q: Why not just use dbt?

A: dbt is excellent for batch transformations in the warehouse, but it cannot serve individual rows to a live API with low latency. Fabra bridges the gap between your dbt models (Offline) and your production API (Online).

Operations

Q: How do I migrate from Local to Production?

A: It's a configuration change, not a code change.

1. Set FABRA_ENV=production.
2. Provide FABRA_POSTGRES_URL and FABRA_REDIS_URL.
3. Deploy. See Local to Production for a guide.

Context Accountability (v1.4+)

Q: How do I track what data my AI used?

A: Fabra automatically tracks lineage for every context assembly. Access via ctx.lineage after calling your @context function, or query historical contexts with store.get_context_at(context_id).

Q: Can I replay an AI decision for compliance or debugging?

A: Yes. Every context gets a UUIDv7 ID. Use store.get_context_at(id) to retrieve the exact content, features, and retriever results that were assembled. This is critical for regulatory compliance — when auditors ask "what did your AI know when it decided?", you have a complete answer.

Q: Where is context lineage stored?

A: In the context_log table in your offline store (DuckDB or Postgres). You can query it directly with SQL or use the CLI: fabra context list.

Freshness SLAs (v1.5+)

Q: How do I ensure my AI context uses fresh data?

A: Add freshness_sla to your @context decorator: @context(store, freshness_sla="5m"). Fabra tracks feature ages and reports violations via ctx.meta["freshness_violations"].

Q: What happens when features are stale?

A: By default (degraded mode), context assembly succeeds but freshness_status becomes "degraded". Use freshness_strict=True to raise FreshnessSLAError instead.

Q: How do I monitor freshness SLA violations?

A: Fabra exposes Prometheus metrics: fabra_context_freshness_status_total (guaranteed/degraded counts), fabra_context_freshness_violations_total (per-feature violations).