Frequently Asked Questions

Real answers to the questions buyers ask before signing off on a custom AI project — on privacy, process, pricing, and everything in between.

How It Works

ChatGPT is a general-purpose AI trained on the public internet. A custom AI chatbot is trained — or more accurately, grounded — on your specific business content: your product catalog, your policy documents, your FAQs, your knowledge base.

This means it can answer questions like "What is the return policy for items bought during sale?" or "Which SKU fits a size M waist of 32 inches?" using your real data — not a generic guess.

The underlying model (GPT-4, Claude, etc.) is still powerful. We just constrain it to your content so it stays accurate and on-brand.

We use a technique called RAG — Retrieval Augmented Generation. Here's the plain-English version:

  1. Your documents (PDFs, Word files, web pages, database records) are split into small chunks and converted into numerical representations called embeddings.
  2. These embeddings are stored in a vector database (like Pinecone or ChromaDB).
  3. When a user asks a question, we search the vector database for the most relevant chunks from your documents.
  4. Those chunks are passed to the LLM (e.g., GPT-4) along with the question. The LLM answers using only that retrieved context.

The result: answers that are factually grounded in your content, not hallucinated from general training data.

RAG Vector Embeddings LLM Grounding No hallucination risk from generic data

We can ingest almost any structured or semi-structured content:

  • Documents: PDF, Word (.docx), Excel, PowerPoint, plain text
  • Web content: Scraped pages, sitemaps, blog posts
  • Databases: MySQL, PostgreSQL, MongoDB — via API or direct connection
  • Product data: CSV exports from Shopify, WooCommerce, custom ERP
  • Internal wikis: Notion, Confluence, SharePoint exports

If your data is in a format not listed here, tell us — we've handled unusual formats before.

Not automatically. The AI retrieves from your indexed knowledge base — it doesn't update itself based on user chats (which would be a security risk if users could inject false information).

What we do build is a feedback loop: unanswered or low-confidence queries are logged so you can review them and update your knowledge base manually or on a scheduled basis. The index is then re-built with the new content.

This is intentional — it keeps you in control of what the AI knows and says.

Data & Privacy

This depends on the hosting model we agree on. We offer three options:

  • Cloud-hosted (default): Your vector index and application run on AWS, Azure, or GCP — in a region of your choice. Queries are sent to OpenAI/Azure OpenAI API with your API key, under their enterprise data handling terms (OpenAI enterprise does not train on API traffic).
  • Your own cloud: We deploy everything inside your existing AWS or Azure account. You own all infrastructure; we have no ongoing access unless you grant it.
  • On-premise / private: For sensitive industries, we can deploy with open-source models (Mistral, Llama) entirely within your servers — no external API calls at all.
We do not store or process your documents on our own servers beyond the initial build phase, unless you explicitly request managed hosting from us.

No — when using the API (not the consumer ChatGPT product), OpenAI does not use your inputs or outputs to train their models. This is covered under their API terms of service and enterprise data processing agreements.

The same applies to Azure OpenAI, Anthropic (Claude), and Google (Gemini via API). All of these providers operate API access under data processing agreements that prohibit training on customer data.

OpenAI API — no training on your data Azure OpenAI — GDPR compliant Anthropic Claude — API DPA available

Yes — and we've built production systems for exactly this scenario (our mylegalware.com product handles confidential trade law cases and investigation files).

For sensitive deployments we implement:

  • Role-based access control — users only retrieve documents they are authorized to see
  • Audit logging — every query and retrieved chunk is logged with user identity and timestamp
  • Encrypted vector store — embeddings stored at rest with AES-256
  • Private model option — on-premise LLM so no data ever leaves your network

You do — fully. On project completion we hand over all source code, vector index configurations, prompt templates, and deployment scripts. There is no lock-in to our infrastructure or our accounts.

You can run it yourself, have another team maintain it, or continue working with us. The choice is always yours.

Integrations

Yes. We deliver the chatbot as a lightweight JavaScript widget — a single script tag you paste into your site's HTML. It works on any platform: plain HTML, WordPress, Shopify, Webflow, custom React/Vue apps.

The widget is fully customizable: brand colors, logo, placeholder text, greeting message, and widget position (bottom-right, bottom-left, or inline).

Average embed time for a developer: under 10 minutes.

Yes. Common CRM integrations we build:

  • Lead capture: When a user provides their email or name during a chat, it's pushed as a new contact or lead to your CRM automatically
  • Context enrichment: If a user is logged in, the chatbot can pull their history from your CRM to personalize responses ("Your last order was X...")
  • Ticket creation: Unresolved queries can auto-create a support ticket in HubSpot, Freshdesk, or Zendesk with the full conversation as context

We use official CRM APIs and OAuth tokens — no credentials are stored in the chatbot code.

Yes — this is one of the most popular use cases for internal knowledge assistants. Employees ask questions directly in Slack or Teams, and the bot answers using your internal documentation, HR policies, onboarding guides, or product specs.

  • Slack: Deployed as a Slack App with slash commands or @mention support
  • Microsoft Teams: Deployed as a Teams Bot via Azure Bot Framework

Access can be restricted by Slack channel or Teams team, so different departments see only their relevant documents.

Yes. We expose all AI functionality via a REST API (or optionally WebSocket for streaming). You can build any interface on top — mobile app, internal portal, voice UI, or integrate it into an existing product.

The API includes endpoints for:

  • Sending a query and receiving an answer with source citations
  • Managing conversation sessions and history
  • Triggering document re-indexing
  • Retrieving usage and confidence metrics
Full API documentation is provided at handover, including Postman collection and code examples in Python, JavaScript, and PHP.

Pricing

We offer two primary models depending on what you need:

  • Fixed-scope project fee: One-time cost to design, build, and deploy your AI system. You own everything. Best if you have an internal team to maintain it afterward.
  • Retainer / product partnership: Monthly engagement where we continue building features, tuning the model, updating the knowledge base, and handling infrastructure. Best for evolving products.

We don't charge a "SaaS seat fee" for your end users — you're not paying per user or per conversation unless you're using a cloud AI API that bills per token (which we help you estimate upfront).

Project: one-time fee Retainer: monthly No per-user seat charges from us

After launch, the main ongoing cost is your LLM API usage (e.g., OpenAI charges per token — roughly per word processed). We estimate this during scoping based on your expected query volume.

As a rough guide:

  • Low volume (100–500 queries/day): typically $20–$80/month in API costs
  • Medium volume (1,000–5,000 queries/day): typically $100–$400/month
  • High volume: we architect with caching and model tiering to keep costs down

You pay these API costs directly to OpenAI/Azure/Anthropic using your own API key — they don't pass through us.

Yes — we offer a paid discovery engagement (1–2 weeks) where we review your data, define the architecture, and often build a working proof of concept on a subset of your documents.

At the end of discovery you get:

  • A technical architecture recommendation
  • A data readiness assessment
  • A realistic scope and cost for the full build
  • Optionally: a working prototype you can demo internally

This is fixed-price and the cost is separate from the full project. If you proceed, the discovery work is credited against the full project fee.

Timelines

A typical project from kickoff to production launch takes 6–8 weeks. Here's how that breaks down:

  • Week 1: Discovery — understand your domain, data sources, and goals
  • Week 2: Architecture design — choose models, vector store, and deployment approach
  • Weeks 3–5: Build — document ingestion pipeline, AI backend, and chat interface
  • Week 6: Testing and tuning — accuracy review, edge case handling, prompt optimization
  • Week 7: Staging deployment and UAT with your team
  • Week 8: Production launch and monitoring setup
Simpler integrations (e.g., adding a RAG layer to an existing app) can ship in 3–4 weeks. Complex multi-module systems (like a full legal research platform) may take 10–12 weeks.

To kick off a project we typically need:

  • Your documents / data: Whatever you want the AI to know — PDFs, spreadsheets, web content, database exports
  • A point of contact: Someone who understands the domain and can review answers for accuracy during testing
  • API access decisions: Which LLM provider you want to use, or we recommend one based on your requirements
  • Brand guidelines (if embedding on website): Logo, colors, tone of voice

We don't need your internal developers to be involved unless you want them to be — we handle everything end-to-end.

Yes — and we build this in from the start. We provide an admin panel where you (or your team) can upload new documents, trigger re-indexing, and see which documents are currently indexed.

Re-indexing after a new document upload typically takes a few minutes, not hours. The chatbot is updated live with no downtime.

Support

All project deliveries include a 30-day post-launch support window at no extra cost. During this time we fix any bugs, tune accuracy issues flagged by real users, and answer technical questions from your team.

After 30 days, you can choose:

  • Self-managed: Take over completely with the codebase and documentation we provide
  • Support retainer: A lightweight monthly arrangement for bug fixes, dependency updates, and occasional tuning
  • Full product partnership: Ongoing feature development and continuous improvement

RAG systems are significantly more accurate than general LLMs for domain-specific questions, but no AI is perfect. We handle this in a few ways:

  • Confidence thresholds: If the retrieval score is low, the chatbot says "I don't have enough information on that" rather than guessing
  • Source citations: Every answer includes a reference to the source document so users (and you) can verify
  • Fallback routing: Low-confidence queries can be routed to a human agent or a "contact us" prompt
  • Query logging: Failed or flagged queries are logged for review so you can improve the knowledge base

During the first 30 days post-launch, we actively review logged queries and tune the system based on real usage patterns.

Yes — we sign NDAs before any document sharing or discovery work begins. For regulated industries or clients handling personal data, we also sign a Data Processing Agreement (DPA) that sets out exactly how your data is handled, stored, and deleted.

For enterprise clients we are happy to review and sign your standard vendor agreements rather than requiring you to use ours.

The fastest way is to fill out our contact form. Describe what you're trying to build, what data you have, and your rough timeline — and we'll respond within one business day.

We also offer a free 30-minute scoping call with no obligation. Use the contact form to request one and we'll schedule at a time that works for you.

We work across time zones and typically respond to enquiries from India, UK, US, and Middle East clients within the same business day.

Still Have a Specific Question?

Tell us about your project. We'll give you a straight answer — no sales pitch, no jargon.

Get in Touch