A private AI assistant that knows your company and keeps your data yours.
A private AI assistant answers employee questions from your own documents, shows its sources, and only returns what each person is allowed to see. We build it with retrieval-augmented generation (RAG), single sign-on and permissions inherited from your systems, deployed in your cloud account so your data stays under your control.
Book a scoping call →What a private AI assistant is, and what it is not
A private AI assistant is an internal tool, similar in feel to a chat assistant, that is connected to your company knowledge: policies, contracts, wikis, tickets, shared drives, product documentation. It retrieves the relevant passages for each question, gives them to a language model as context and answers with citations back to the source. It is not a public chatbot with your files pasted into a prompt, and it is not a model trained on your data. Your documents stay in an index that you own. That distinction matters for security reviews: you can point to where the data lives, who can read it and how to delete it.
How it works: RAG, SSO and permissions
The system has four layers. Connectors pull content from the sources you choose and keep it in sync. An index splits documents into passages and stores them for search. A retrieval step finds the passages relevant to a question, filtered by what the asking user is allowed to read. A language model then writes an answer using only those passages and cites them. Sign-in goes through your identity provider with SSO, and document permissions are carried over from the source system, so a salesperson does not receive an answer built from an HR file. When the assistant cannot find support for an answer, it says so instead of guessing.
| Layer | What it does | What you control |
|---|---|---|
| Connectors | Sync documents from drives, wikis, ticketing and databases | Which sources are included and how often they refresh |
| Index | Stores passages and metadata for search | Where it is hosted, retention and deletion |
| Retrieval with permissions | Finds relevant passages the user may read | Access rules inherited from the source system |
| Model and answer | Writes a cited answer from retrieved passages | Model provider, region and logging settings |
| SSO and audit | Signs users in and records questions and sources | Identity provider, roles and log retention |
Where your data stays
We deploy the assistant in your cloud account, with your repositories and your keys, so documents and the index do not sit in an environment we control. The model is called through a provider configured under your account and settings, and we help you choose one whose data-handling terms match your policy, including whether prompts are retained or used for training. If your policy requires a model running inside your own network, we scope that explicitly, because it changes cost and quality. We do not claim any security certification for this system. What we do provide is an architecture description, an access model and logs that your security team can review before launch.
Where a private assistant pays off first
The best first use cases are narrow and document-heavy: an internal policy and HR helper, a support team searching past tickets and product documentation, a sales team finding contract terms, an operations team following standard procedures. A narrow scope makes the answers testable. Before launch we build an evaluation set of real questions with known answers and measure the assistant against it, and after launch AgentOps from $1,500 per month keeps watching quality as your documents change. Our AI Integration and RAG service starts at $15,000, and an AI Readiness Sprint from $4,500 can first decide which use case and which data are worth starting with.
- Internal policy, HR and IT helper
- Support search across tickets and product documentation
- Contract and proposal lookup for sales and legal operations
- Standard operating procedures for field and operations teams
Frequently asked questions
Is a private AI assistant the same as ChatGPT for my company?
No. A general chat tool answers from public training data. A private assistant answers from your own documents, cites them, and applies your permissions. It runs in your cloud account with your identity provider.
Will it train a model on our data?
No. We use retrieval, not training: your documents stay in an index you own and are passed to the model as context for each question. We help you choose a model provider whose terms on retention and training match your policy.
How does it respect who can see what?
Permissions are inherited from the source systems and applied at retrieval time, so the assistant only builds answers from passages the signed-in user may read.
What does it cost?
AI Integration and RAG starts at $15,000, an AI Readiness Sprint starts at $4,500, and AgentOps for monitoring after launch starts at $1,500 per month. The final price depends on your sources and permission model.
Who owns the code?
You do. Clients own 100% of the code and IP once each milestone is paid, and the code lives in your repositories from day one.