# An enterprise chatbot that respects your permissions, your systems and your audit trail

An enterprise chatbot is a conversational assistant connected to company systems such as ERP, CRM and HR software, where every answer is limited to what the person asking is allowed to see. Goldenroz builds custom enterprise chatbots with single sign-on, role-based permissions, audit logs and controlled access to on-premises data through MCP. Projects start at $25,000, fixed price per phase after a free one-hour scoping call.

**From $25,000, fixed price per phase**

Updated: 2026-11-14. By Shai Goldenberg.

## What usually goes wrong

- **A pilot that security will not approve:** A team built a demo with one service account and a database password. It would show payroll to anyone who asked, and the security review stopped it.
- **The data sits behind the firewall:** The ERP and the SQL servers have no public API. Hosted chatbot products can read a help center and cannot reach the systems where the real answers live.
- **No record of who asked what:** Staff already paste exports into public AI tools. Compliance cannot say which records left the company or who sent them.
- **Per-seat platform pricing across the whole company:** The platform quote multiplies a monthly seat price by every employee, including the many who would use the bot twice a month.

## What is included

- **Single sign-on:** Login through your identity provider using standard protocols such as SAML or OpenID Connect, so access follows your joiner and leaver process.
- **Role-based permissions:** Each request carries the user's identity. The bot returns only what that person may already see in the source system, enforced in the data layer and not in the prompt.
- **System integration:** Connections to ERP, CRM, HRIS, ticketing and internal databases through their APIs or approved read-only queries. See [AI integration](https://goldenroz.com/ai-integration/).
- **On-premises access through MCP:** A gateway inside your network that exposes approved queries to the model, with personal fields masked before results leave. Covered on the [MCP integration](https://goldenroz.com/mcp-server-development/) page.
- **Audit log:** Every question, every query run, every record returned and every model call, stored in your own environment.
- **Security review support:** Architecture and data-flow documents, a list of subprocessors, and answers to your security questionnaire.

## Enterprise chatbot platform or custom build

An enterprise chatbot platform gives you a builder, connectors and an admin console for a subscription. It is the right buy when your use case matches its connectors, your data is in cloud tools it already supports, and seat pricing works at your headcount.

A custom build fits in three situations. The data is on-premises or in a system with no connector. The permission model is more detailed than the platform's roles. Or the bot has to live inside your own product or portal, with your interface. In those cases a custom build leaves you owning the code, the prompts and the logs.

## An LLM chatbot and where your data goes

An LLM chatbot sends text to a model to get an answer, so the first enterprise question is which text and to whom. In our design the databases stay where they are. Only the results of approved queries, after masking, go to the model as context, under your own agreement with the model provider.

Model choice is yours. Claude is our default, OpenAI models are supported, and an open-source model hosted in your own environment is an option when no data may leave it. The bot is built so the model can be replaced without rewriting the integrations.

Data residency depends on the regions your model provider and cloud offer and on your contract with them. Goldenroz is not a compliance auditor and holds no partner status with Anthropic, Microsoft or OpenAI.

## What to ask a conversational AI company before you sign

Ask any conversational AI company the same five questions. Who writes the code, and can you meet them. Who owns the code and the conversation logs. How a user's permissions are enforced. What happens when the model is down. What the bot is forbidden to do.

Our answers: [Shai](https://goldenroz.com/about/shai/), our CTO, builds it. You own everything. Permissions are enforced where the data is read. A second model and fixed replies cover outages. The forbidden list is written at scoping. Shai built an on-premises MCP gateway of about 27,000 lines with 13 connector types, read-only access, per-user permissions, PII masking and an audit log.

## What it costs

Each phase has a fixed price agreed before work starts. Model usage and WhatsApp messaging are billed by the providers to your own accounts.

- Support or ordering bot: From $25,000 (Fixed price per phase)
- Multi-channel or multi-brand platforms: $100,000 and up (Scoped per project)
- Scoping call: $0 (Ends with a written scope and a price)
- Model and WhatsApp usage: Paid to the providers (Billed to your own accounts)

## Who it is not for

- Giving a model a database password and free-form queries. We do not build that.
- Write access to core systems on day one. Integrations start read-only and widen one action at a time.

## Questions buyers ask

### How much does an enterprise chatbot cost to build?

Projects start at $25,000, fixed price per phase after a free one-hour scoping call. Programs across several departments and many systems run $100,000 and up. Model usage is paid to the provider from your own account.

### Can the chatbot read data from our on-premises ERP?

Yes, through a gateway installed inside your network. It runs approved read-only queries under the identity of the user asking and masks personal fields before results reach the model. The model never writes SQL and never holds a database password.

### Do you offer conversational AI services beyond the first build?

Yes. After launch, new use cases and systems are added as further fixed-price phases.

### Is a custom chatbot for business safer than a public AI tool?

It can be, when it is built with per-user permissions, masking and logging. A public tool has no link to your access rules and no record of what staff pasted in.

### Are you a chatbot agency or an engineering firm?

We are a small engineering firm. A chatbot agency usually writes conversation scripts on a packaged tool, and that is not our work. We build the software, the integrations and the controls, and a named engineer does it.

---
Page: https://goldenroz.com/enterprise-chatbot-development/
Book a free one-hour call: https://calendly.com/daniarozin/daniel-rozin-consultation-1-hour-clone
