An AI assistant that answers from your company's documents and shows where it found the answer
An AI knowledge assistant is a private chatbot that answers staff questions from your own documents, wikis, tickets and shared drives. It uses retrieval-augmented generation (RAG) to find the relevant passages and cite them, answers only from sources your company controls, and only from documents the person asking is allowed to see. Sysvoy builds knowledge assistants that connect to sources such as SharePoint, Google Drive, Confluence, Notion and help desks, run on the model and cloud you choose, and are tested against real staff questions before launch. Most cost $8,000–$20,000 for one main source and $20,000–$45,000 across several sources with permissions and monitoring.
The path a question takes
Documents are parsed and split, embedded, searched, and the answer cites the passages it came from.
- Ingest
- Chunk
- Embed
- Retrieve
- Answer
Where internal AI assistants stall
And what we do differently.
Stale and duplicate documents
Three versions of the travel policy produce three different answers. We pick a source of truth for each topic, drop duplicates, and sync deletions as well as changes.
Permissions added at the end
An assistant that can read everything will eventually quote the salary spreadsheet. We apply each source's access rules at search time from the first prototype.
Answers nobody trusts
Without sources, people check every answer themselves and stop using the tool. Ours cite the passage for each claim and say so when the documents don't cover a question.
What we build
Policy and HR assistants
Answers on leave, expenses, benefits and procedures from your handbook and policies, with a link to the exact clause.
Support and operations assistants
Answers from runbooks, past tickets and product docs, so support staff stop asking engineers the same questions.
Sales and proposal assistants
Finds the right case study, security answer or past proposal paragraph, and drafts from it with sources attached.
Assistants in Slack and Microsoft Teams
The assistant lives where people already ask questions, and posts its sources with every answer.
Self-hosted assistants
Open-weight models and search running in your cloud or on your servers, when documents can't leave your environment.
How we deliver
From the first workshop to production. How we work
Start from real questions
We collect 50–100 questions your staff actually ask, with the documents that answer them. That set decides scope and becomes the test.
Connect and keep sources in sync
We parse documents properly, including tables and scans, and re-index changes and deletions, so retired policies drop out of answers.
Permissions from day one
Access rules come across from each source and are applied at search time, so the model never sees a passage the user couldn't open themselves.
Measure, launch, improve
We launch to one team once the test set passes, then use feedback and the log of unanswered questions to fix gaps in retrieval and in the documents.
A custom knowledge assistant compared with off-the-shelf AI assistants and enterprise search
| Off-the-shelf AI assistant | Enterprise search | Custom knowledge assistant | |
|---|---|---|---|
| Pricing | Per user, per month | Per user or per index, usually annual | One-off build, plus usage at cost |
| Sources | Strongest inside its own suite | Many ready-made connectors | Any source with an API or export |
| Answer rules | Limited control | Returns documents, little drafting | Yours: what to answer, what to refuse, when to hand off |
| Hosting and model | Vendor's cloud and models | Vendor's cloud | Your cloud or servers, model of your choice |
| Best for | Everyday writing and email across the suite | Finding documents across many tools | Accurate answers on a defined body of knowledge |
A good fit when
- The same questions reach HR, IT or senior staff every week, and the answers already exist in a document.
- New hires take months to learn where things are and who to ask.
- You tried an off-the-shelf AI assistant and it can't see the right sources, or sees too many.
What we build with
- Python
- TypeScript
- PostgreSQL and pgvector
- Typesense
- SharePoint, Google Drive, Confluence and Notion APIs
- Slack and Microsoft Teams
- OCR and table parsing
- OpenAI, Claude and open-weight models
- Langfuse
- AWS and Google Cloud
Related work
midpage.ai
AI research platform that layers analysis over case law for litigators and legal teams.
midpage.ai ↗AI commerce assistant
Self-hosted AI shopping assistant, Typesense product search and HubSpot integration for a US B2B distributor.
US distributor
Common questions
Not answered here? Email contact@sysvoy.com.
How much does an AI knowledge assistant cost?
Most cost $8,000–$20,000 for one main source, such as a SharePoint site or a help desk, and take 4–8 weeks. Assistants across several sources, with per-user permissions, monitoring and admin tools, usually cost $20,000–$45,000 and take 8–14 weeks. Running costs for models and hosting are typically $100–$2,000 a month, paid at cost. A 1–2 week discovery sprint turns this into a fixed quote.
How is this different from Microsoft 365 Copilot or ChatGPT Enterprise?
Off-the-shelf assistants charge per user each month (Microsoft 365 Copilot lists at $30 per user per month for enterprise plans) and work best on content inside their own suite. A custom assistant costs a one-off build plus usage, answers from exactly the sources you choose, follows your rules on what to answer and what to refuse, and can run on your own cloud and model. Many companies use both: a general assistant for everyday writing and a custom one for the knowledge that has to be right.
Will people only see answers from documents they're allowed to open?
Yes. We copy access rules from each source, such as SharePoint groups or Google Drive sharing, and filter search results by the user's permissions before anything reaches the model. Before launch we test with accounts for each role to confirm that restricted documents never appear in answers or citations.
What happens when the answer isn't in the documents?
The assistant says it couldn't find an answer and, where you want it to, points to the team that owns the topic. It doesn't fall back on general knowledge for company questions. Unanswered questions are logged, which gives you a running list of gaps in your documentation.
Which sources can it connect to?
SharePoint and OneDrive, Google Drive, Confluence, Notion, help desks such as Zendesk and Jira Service Management, databases, and folders of PDFs, Word files and scans. If a system has an API or an export, we can index it and keep it in sync.
How do you know the answers are accurate?
We measure them. Before launch, the assistant is tested on a set of real staff questions with known answers and sources, and we report how many it answers correctly with the right citation. We agree the pass mark with you, and rerun the same test after every change to documents, prompts or models.
Can it run on our own servers?
Yes. We deploy to your cloud account or your own servers. Where documents can't be sent to an outside model provider, we use open-weight models served inside your environment, so nothing leaves it.
Have something to build?
Tell us where things stand and what you need. A short message is enough to start.