RAG development services
RAG development: an AI chatbot for your company documents that answers with citations
We build private AI assistants that read your Google Drive, Notion, SharePoint, Confluence and PDFs. Staff ask in Slack, Teams or a web app and get an answer that links to the source document. Each person sees only what their role allows.
- Answers link to the source document — and to the section where there is one
- Role-based access and an audit log — private and secure by design
- Evals and an accuracy report on your real questions — from $2,500, live in 2–3 weeks
Or message us directly
Tell us about your documents — plan and fixed quote in 24 hours
Where your docs live (Drive, Notion, SharePoint, Confluence, PDFs), who will use the assistant and a few questions they ask today.
- 24 hto a plan and fixed quote
- 2–3 weekstypical time to launch
- 30 daysof tuning included
- 1 engineerowns your project end-to-end
See it work
Tuesday, 9:41 am. A travel question in Slack, answered from the current policy with the section cited
Here’s an internal knowledge base AI at work in Slack. It checks who is asking, searches only the documents that role can open, answers from the latest version and routes the question to the right owner when the answer isn’t there.
Answers show their source
The document name and a link, plus the section when the file has one. People check the source in one click instead of taking the bot’s word for it.
Access follows roles
The assistant checks who is asking before it searches. Finance, HR and legal files stay with the people allowed to see them.
Old versions stay out
Documents you mark as archived or superseded are skipped, so answers follow the current version. Questions with no answer go to the document owner.
Hey! Can I book premium economy for the Lisbon trade show next month? The flight is about 7 hours. Does anyone need to sign off?
access.check(user="j.alvarez", role="Account Manager") → HR & Travel: allowed · Finance: deniedkb.search("premium economy long-haul approval", sources=["Drive/Policies", "Notion/Ops"]) → 4 passages · Travel Policy v3.2 (current) · v2.0 (archived, skipped)Yes. Premium economy is allowed on flights over 6 hours, but your manager needs to approve it before you book. Heads-up: this changed in March — the old 8-hour rule no longer applies. 📄 Travel & Expense Policy v3.2, §4.1 and §4.3
Perfect, thanks. And how much of the sales travel budget is left this quarter?
kb.search("sales travel budget Q3", sources=allowed_for("Account Manager")) → 0 passages · routing rule: budget questions → #finance-requestsI couldn’t find that in the documents your role can open, so I won’t guess. Budget questions go to Finance — want me to post it in #finance-requests for you?
Yes please.
- Request posted to #finance-requests · question, sources and answer saved to the audit log
Demo scenario: the company, people and documents are fictional.
What it does
Where an internal knowledge base AI saves your team time
Your team already asks these questions in Slack, Teams and email every day. The assistant answers them from your documents and shows where each answer came from.
- 01
HR, IT and ops questions
Leave, travel, expenses, laptop setup, who approves what. Staff chat with company documents in Slack or Teams and get an answer that cites the policy section.
- 02
Onboarding new hires
New people ask the assistant instead of pulling senior colleagues away from their work. It points to the right template, process doc or owner.
- 03
Contract and SOP search
Finds payment terms, SLAs and renewal clauses in contracts and PDFs and shows the exact clause. Legal judgment stays with your lawyer.
- 04
Answers for sales and support reps
Reps check warranty terms, shipping rules or product specs mid-conversation and give the same answer as everyone else on the team.
- 05
Custom knowledge base chatbot for customers
Reuse the same knowledge base for a website or WhatsApp bot. It sees only public docs and hands tricky questions to a person with a summary.
- 06
MCP and Claude for internal tools
We build MCP servers so Claude or your own agents can search your docs, tickets or database — with the same access rules and logging.
- 07
Gap log for content owners
See what people ask most and where the docs have no answer, so owners know which pages to write or update first.
What we build
Knowledge-base AI, support answers and document workflows
Start with an internal assistant on your documents. Reuse the same knowledge base for a customer support agent, and add n8n workflows that keep sources in sync.
01 / KNOWLEDGE
Knowledge-base assistant (RAG)
An internal assistant that knows your SOPs, contracts and docs — and cites them.
- Answers staff questions with citations
- Google Drive, Notion, SharePoint, PDFs
- Role-based access and audit log
02 / SUPPORT
AI customer support agent
Resolves repetitive tickets instantly and escalates the rest with full context.
- Answers from your help center with sources
- Order status, returns, account questions
- Hands off to Zendesk, Intercom or Slack
03 / AUTOMATION
AI workflow automation
n8n and Make workflows with AI steps that remove manual work between your tools.
- Lead routing, enrichment and follow-ups
- Document and email processing
- Self-hosted n8n — no SaaS lock-in
Need something bigger?
Multi-step agents, MCP servers and Claude integrations for internal tools, company-wide AI assistants with human approvals and access control — designed around your systems. White-label builds for agencies, too.
Private and secure by design
Built so your security team can review how it works
The worries we hear most about a chatbot for company documents: it will leak files, make things up, or lock you into a vendor. Here is how we handle each one.
Business APIs that don’t train on your data
We use model providers’ business APIs, which by default don’t train on your prompts or files. The accounts are in your name, so the data terms are between you and the provider — and you can check them yourself.
Role-based access control
Permissions are checked before every search. Where possible, roles mirror your groups in Google Workspace or Microsoft 365.
Audit log of every answer
Each question, the passages used and the answer are logged. You can see who asked what and why the assistant replied the way it did.
Evals and accuracy reports
Before launch, we test the assistant on your real questions against thresholds agreed in writing. If it falls short, we fix it at no cost.
Self-hosting option
Need data to stay in your own cloud? We can deploy the search index and the app on your infrastructure and pick models that fit your rules.
You own the whole stack
Code, index, prompts, test set and accounts are yours. Model and hosting usage is billed at cost, with no markups.
Compare
Shared-drive search, built-in AI, a DIY chatbot or a custom RAG assistant?
If all your docs live in one app and permissions are simple, the AI built into that app may be enough. A custom assistant pays off when knowledge is spread across tools, access differs by role, or answers must be checked.
| — | Search the shared drive | AI built into one app | DIY no-code chatbot | Custom RAG assistant |
|---|---|---|---|---|
| Where it looks | File names and keywords | Mostly inside that app | Files you upload to it | Drive, Notion, SharePoint, Confluence and PDFs together |
| What you get back | A list of files to open | An answer; citations vary by tool | Depends on your setup | A direct answer that cites the source document |
| Access control | Folder permissions | That app’s permissions | Depends on the builder | Checked by role before every search |
| Checked before launch | — | You test it yourself | You build the tests | Evals on your real questions, thresholds agreed in writing |
| Where people ask | In the drive | In that app | A widget you host | Slack, Teams, web, Telegram or via MCP |
| Costs | Staff time spent searching | Usually a per-seat subscription | Platform plan plus your time | Build from $2,500, usage at cost on your accounts, optional care |
How we work
From a folder of documents to a tested assistant — usually in 2–3 weeks
Day 0
Brief and NDA
Tell us where your documents live, who will ask questions and what they ask today. We sign an NDA before you share files. Within 24 hours you get a plan, stack and fixed quote.
Days 1–5
Test set and prototype
You send a list of real questions your team asks, with the answers you expect. We build a prototype on a sample of your documents and show cited answers before the main invoice.
Weeks 1–3
Sources, roles and evals
We connect Drive, Notion, SharePoint or Confluence, set up roles and channels, then run the test set. You get an accuracy report against the thresholds we agreed.
After launch
Sync and care
Updated documents are re-indexed on a schedule. 30 days of tuning are included, then optional care: gap-log reviews, new sources and test-set re-runs.
Pricing
RAG chatbot pricing: fixed build, model usage at cost
A knowledge-base assistant is quoted on its own: from $2,500, usually live in 2–3 weeks, at a fixed price based on your sources, roles and channels. The plans below are our general agent packages. Starter covers a simpler customer-facing FAQ bot, not an internal assistant with roles and citations. Company-wide setups with several teams, MCP integrations and approvals fit Custom.
Starter
One agent, one channel — chat or voice
from $2,000
care from $250/molive in 1–2 weeks
- Website chat, WhatsApp or phone
- Knowledge base: pricing, FAQ, policies
- Lead capture and booking into 1 calendar
- Human hand-off and summaries
- 30 days of tuning
- Most popular
Growth
Omnichannel sales & booking with CRM
from $3,500
care from $450/molive in 2–3 weeks
- Voice + chat on up to 3 channels
- CRM integration (HubSpot, GHL, Pipedrive…)
- Follow-ups, reminders, missed-call text-back
- Tested on your real conversations
- Dashboard and weekly reports
- 30 days of tuning
Custom
Complex workflows and internal AI
from $8,000
care from $900/molive in 3–6 weeks
- Multi-agent workflows with approvals
- RAG with citations and access control
- MCP / Claude integrations to your tools
- Evals and accuracy reports
- Security review, docs and team training
- Milestone payments — the larger part only after acceptance against the agreed test set.
- Model, search and hosting usage is billed at cost on your own accounts. We estimate it before we start.
- Already have a RAG chatbot that gives wrong answers? Rescue & audit from $400.
FAQ
Questions about RAG chatbots for company documents
How much does RAG chatbot development cost?
A knowledge-base assistant is a fixed-price build that starts at $2,500 and usually goes live in 2–3 weeks. The quote depends on how many sources, roles and channels you need. Model and hosting usage is billed at cost on your own accounts. 30 days of tuning are included, and an optional care plan is priced in the same quote.
What is a RAG chatbot, in plain English?
A RAG chatbot is a chatbot with a knowledge base behind it. RAG stands for retrieval-augmented generation: before it answers, the AI searches your documents for the relevant passages and writes the answer from them. That’s why it answers from your policies instead of giving generic internet advice, and why it can show the source.
Can you build a chatbot trained on our own documents?
Yes, though usually without training the model itself. The assistant searches your documents each time someone asks, so an updated policy shows up in answers after the next sync. That’s easier to keep current and to control than fine-tuning.
Can it connect to Google Drive, Notion, SharePoint and Confluence?
Yes. These are the most common sources, and we connect them through their APIs with read access you grant. PDFs, Word files and spreadsheets work too, and help centers, ticket systems, CRMs or databases can be added via API. Scanned files work after OCR, and cleaner scans give better answers.
Is it safe to give an AI access to company documents?
It reads only the sources and folders you connect, and it checks each person’s role before it searches, so files outside that role stay out of the answer. Model calls go through business APIs that by default don’t train on your data, on accounts in your name. Every question, source and answer is logged, and the search index and app can run in your own cloud if your rules require it.
How do you measure answer accuracy before launch?
We collect real questions from your team, the answers you expect and the cases that must go to a person. We agree acceptance thresholds in writing and run the assistant against that set. You get a report with every answer, source and miss. After launch, re-runs are part of the 30-day tuning or an optional care plan.
What happens when the answer isn’t in our documents?
The assistant is set up to say it couldn’t find the answer rather than guess, and to route the question to the owner you choose. These questions go into a gap log, so you can see which docs to write or update next.
Let’s start
See cited answers from your own documents before you pay the main invoice
Tell us where your docs live and share a few real questions your team asks. We sign the NDA first. Within 24 hours you get a plan and a fixed quote — then you test a prototype on your own documents.
Or message us directly