What document formats do you support for RAG ingestion?+
PDF, Word, Excel, PowerPoint, HTML, Markdown, plain text, and most common formats. Codefree can also extract from web pages, Confluence, Notion, SharePoint, and REST APIs — the ingestion pipeline is format-aware and handles the parsing per format, including tables, images with OCR, and structured metadata.
How do you handle document updates in the RAG system?+
Codefree sets up automatic re-indexing pipelines that detect changes and update the vector store, keeping your knowledge base current. Depending on your update frequency, we can schedule re-indexing hourly, daily, or on document-change webhooks from Confluence, SharePoint, or your CMS.
Can users see where the AI answers come from?+
Yes — every response includes source citations with links back to the original documents and the specific relevant passages. Users can click through to verify the source in seconds, which is what makes RAG-based answers trustworthy compared to generic ChatGPT that cannot cite anything.
How accurate is the system in production?+
RAG dramatically reduces hallucinations by grounding every answer in your actual documents. Codefree also implements confidence scoring so low-confidence answers can be flagged for human review or routed to a human agent instead of guessing. Typical production accuracy for well-scoped RAG deployments is 85 to 95 percent, with the remaining edge cases handled by human-in-loop escalation.
What is the real total cost — model fees, hosting, ongoing?+
RAG typically runs 30 to 300 USD per month depending on query volume (embedding, LLM calls, and vector database costs). Multi-Agent runs 200 to 2,000 USD per month based on orchestration complexity. Fine-tuned models are typically self-hosted with infrastructure costs starting at 200 USD per month. Codefree gives you a 12-month cost projection before kickoff so there are no surprises after go-live.
What if our data is not ready or is not clean?+
Data preparation is part of every plan — Codefree does not expect pristine inputs. We show you a data quality report after the first week and recommend either proceeding, fixing specific gaps first, or scoping a separate data-cleanup engagement. We will not quietly ship something that performs badly because the data was not ready.
Are we locked into a specific AI vendor like OpenAI or Anthropic?+
No — the system is designed to be portable across providers. Codefree typically builds with LangChain abstractions over OpenAI, Anthropic, and open-source models (Llama, Mistral, Qwen). You can switch providers or self-host without rebuilding. The architecture and prompts are yours to keep and modify.
What if the AI model we are using gets deprecated?+
Major models (OpenAI GPT-4, Anthropic Claude) stay available for 12 to 18 months after deprecation announcements, with clear migration paths. Codefree's builds version-pin the model and document upgrade steps. The 30-day post-launch support window includes a no-cost model upgrade if a deprecation lands in that window.
Who owns the trained model or the RAG system?+
You own the application, the prompts, the data pipelines, and (for Fine-Tuning Enterprise) the model weights outright. Nothing is built on a Codefree-controlled platform you cannot take with you — everything ships to your infrastructure or your cloud account with source code on your GitHub.
Which plan should I choose — RAG Starter, Multi-Agent Growth, or Fine-Tuning Enterprise?+
RAG Starter is best for straightforward Q&A over documents (support bots, knowledge bases, product docs). Multi-Agent Growth handles complex workflows requiring multiple specialised perspectives or decision steps (research, compliance review, customer triage). Fine-Tuning Enterprise is for when you need a model that natively understands highly specialised terminology or proprietary domain language (legal, medical, insurance). Codefree will tell you straight on the kickoff call if you have picked the wrong one.
What happens during the 2 to 4 week build timeline?+
Week 1: Discovery call, document audit, and architecture planning. Week 2: Document ingestion and vector store indexing on Pinecone or Weaviate. Week 3: RAG pipeline development, chunking optimisation, and retrieval tuning. Week 4: Testing, edge case handling, and production deployment with handoff. Multi-Agent Growth and Fine-Tuning Enterprise extend into a fifth week for orchestration or model training.
How does the deposit and payment structure work?+
A 50 percent deposit is required to begin work and secure your kickoff slot. The remaining 50 percent is due on delivery and acceptance of the system. This is a one-time engagement with milestone-based payments available for Multi-Agent Growth and Fine-Tuning Enterprise tiers. Payment options: Stripe (credit card), Razorpay, invoice, or wire transfer.
Can this integrate with our existing tools like Slack, Zendesk, or an internal portal?+
The package includes a query REST API and admin dashboard, so you can integrate it into Slack, Zendesk, Intercom, or internal tools via the API. Custom user-facing UI beyond the admin dashboard is not included in the base scope — the Custom Query UI Build add-on covers Slack bots, Teams bots, or branded web widgets.
What does the 30-day post-launch support actually cover?+
Bug fixes, performance tuning, and questions about operating the system Codefree delivered. It does not cover new features, additional document sources, scope changes, or ongoing hosting costs — those are separate. You also get 2 rounds of revisions during the build itself before Day 20 handover.
What if we only have 500 documents right now but expect to grow?+
The package supports up to 10,000 documents out of the box, so you are covered for growth. If you exceed that or need a different tier later, Codefree can discuss scaling the vector store and ingestion pipeline — but you will not hit limits at 500 or 5,000 documents.
Is this a good fit if we do not have documentation yet, just internal expertise?+
No — Codefree needs existing documented knowledge to work with (PDFs, Word docs, Confluence, web pages, structured data). If your knowledge only exists in people's heads, you will need to document it first or consider a different approach. Codefree makes AI use your knowledge; we do not create the knowledge for you.
Can we switch between OpenAI, Anthropic, or open-source models later?+
Yes — Codefree builds with LangChain abstraction layers so you are not hard-locked to one provider. Switching embeddings or LLMs requires some re-indexing and testing, but the architecture supports it. Fine-Tuning Enterprise can also deliver self-hostable model artefacts if you want full infrastructure control.