RAG-in-a-Box.

Three options. (1) Start with your most-queried 10K and add more later — most clients find 5,000-8,000 high-value documents serve 80%+ of real queries. (2) Add additional document tiers at $2,500 per 5,000 docs as a change order. (3) Move to a Studios Growth engagement ($35K-$75K) for unlimited document volume, permission-aware retrieval, and multi-source pipelines. The 10K limit isn't a technical ceiling — it's the scope boundary for fixed-price delivery.

All standard formats: PDF (including scanned with OCR), Word (.docx), PowerPoint (.pptx), Excel (.xlsx with description), Markdown, HTML, plain text, RTF. Source-system native formats: Google Docs/Sheets/Slides, Notion pages, Confluence pages, SharePoint documents, Box files, Dropbox files. Code repositories with documentation. For unusual formats (proprietary CAD files with descriptions, legal-specific formats), we'll confirm during the intake call. If your content is structured data (database records, API logs), that's a different pattern and may need Studios Growth scoping.

Well-built RAG dramatically reduces hallucinations but doesn't eliminate them. Three controls built in by default: (1) the model is instructed to cite sources and abstain when no relevant content is retrieved; (2) the eval suite we build in week 1 scores every retrieval against your test cases — you see the actual accuracy number on your content before launch; (3) low-confidence answers display a "low retrieval confidence" notice instead of guessing. Typical accuracy on well-curated corpora: 88-95% on the eval set. Accuracy depends heavily on document quality and clarity — we'll be honest in week 1 if your content needs cleanup before indexing will work well.

Your documents live in your chosen vector database (Pinecone, pgvector, Qdrant, Weaviate — recommended in week 2 based on your data and infrastructure). At query time, only the small retrieved chunks relevant to the specific question are sent to the model API for synthesis — never your full document set. We default to providers with enterprise terms guaranteeing no training on API inputs (Anthropic, OpenAI Enterprise). For maximum data sovereignty, the entire stack including generation can run on open-source models in your infrastructure — that may push timeline by 1-2 weeks but stays within the productized price. Details in our Privacy Policy.

Three components, all pass-through (we never mark up): (1) Vector DB hosting — typically $70-$400/month for under 10K docs depending on provider. (2) Embedding regeneration — usually under $100/month for steady-state operation, more during initial indexing. (3) Generation API costs — heavily query-volume dependent, typically $100-$800/month for internal teams under 100 users. We model expected costs in week 2 and build a cost dashboard so you can monitor in real time. Optional Studios retainer ($5K-$25K/month) covers ongoing optimization and feature additions for clients who want active management.

Yes, that's the refresh pipeline. We default to nightly reindexing of changed documents from your source systems. For systems with webhook support (Notion, Google Drive, Confluence), we can move to near-real-time updates — typically 5-15 minute lag. Stale content is the #1 RAG quality killer, so the refresh pipeline is core scope, not an upsell. You can also manually trigger reindex from the admin panel whenever you want.

Yes — customer-facing doc search is one of the most common RAG-in-a-Box deployments. Two notes for customer-facing use: (1) we recommend an additional eval pass with adversarial queries (people will ask the system everything, including out-of-scope questions) — we include this for customer-facing deployments at no extra cost. (2) We strongly recommend adding rate-limiting and basic monitoring for cost control on consumer traffic — included by default. For embed-in-your-site widgets (a chatbot bubble), we deliver the embed code as part of the week 5 handoff.

RAG-in-a-Box includes single-tier authentication — logged-in users see all indexed documents, public users see only what you mark public. For more granular permission-aware retrieval — where different users see different documents based on their existing access in source systems — that's part of Studios Growth scope (typically $35K-$75K). If you have a mix of public help-center content and internal-only content, two RAG-in-a-Box deployments (one public, one authenticated) is often cheaper than one Growth engagement.

You own the custom code, custom configurations, and indexed knowledge base — assigned to you on full payment. We retain ownership of our methodologies and reusable RAG patterns (our "Background IP"), but those are licensed to you perpetually as embodied in your deliverables. The source code is in your repo. The vector DB is on your account. There's no lock-in. If you want to take this to another vendor or your internal team in six months, everything you need is in the handoff package. Full IP terms in our Terms of Service §8.

The 50% deposit is non-refundable once we start week 1 work. If we materially fail to deliver against the eval set we agreed on in week 1, we work to fix it under our limited services warranty — re-perform or correct, that's covered in Terms §11. If you terminate without cause mid-engagement, you owe for work performed through termination. We're confident enough in fixed-scope productized delivery that we publish prices and timelines on the website — the warranty terms reflect that confidence.