--- title: "AssistantGeneral FAQ" description: "Is it really private, does it work offline, which models it supports, how the knowledge base works, and what each edition includes. Answered plainly." url: "https://royalsoftworks.com/products/assistant-general/faq/" source: "https://royalsoftworks.com" format: "markdown" note: "Markdown rendering of the HTML page at `url`. Same content, same canonical URL." --- FAQ # Questions we hear most. Does it need internet? Which AI does it use? Is it really private? Straight answers, no sales talk. 01 Is AssistantGeneral actually private? Yes — by architecture. The default model runs locally via llama.cpp, your documents are indexed on your own machine, and nothing has to leave it. Cloud model providers are entirely optional and opt-in, per task. There is no mandatory telemetry and no background calls home. 02 Does it require an internet connection? No. After installation everything works offline — local inference, the knowledge base, workflows, voice and the agent loop. It is designed for air-gapped and network-restricted environments. You only need a connection if you choose to use a cloud model provider or sync an online data source. 03 Which AI models does it support? Any GGUF model compatible with llama.cpp runs locally — Llama, Mistral, Qwen, Phi, Gemma, DeepSeek and hundreds more. You can also connect cloud providers in one click: Anthropic Claude, OpenAI, Google Gemini, xAI Grok, Groq, Mistral, DeepSeek, Together AI, OpenRouter, or any OpenAI-compatible endpoint. Several machines or GPUs can be pooled into one routed cluster. 04 Is it a chatbot or an agent? An agent. Every request runs through a pipeline that plans the work, retrieves relevant documents, calls tools, and then runs a review and a groundedness check before answering. It tracks a task plan, and an inference trace lens lets you see every step and tool call it made. 05 Can it produce real documents and spreadsheets? Yes. It generates genuine .xlsx and .docx files by running Python in a sandboxed runtime, then re-opens each file, audits it for errors, placeholders and broken references, and self-heals before handing it over. You get a finished workbook with computed figures — not a description of one. 06 How does the knowledge base work? Documents are indexed locally in a vector store. Retrieval combines semantic vector search and keyword search, fused with Reciprocal Rank Fusion and then reranked for precision, with diversity, de-duplication and recency weighting applied. It indexes 8+ formats including PDF (with scanned-page text extraction), Office files, HTML and images, and you can keep multiple separate knowledge bases. 07 Can I see what's inside my knowledge base? Yes. An Insights view maps the entire knowledge base visually: documents grouped into automatically labeled clusters with links between related clusters, a weighted topic cloud, a timeline of documents by year, a network of the people and organisations that appear together, and an entity graph showing which documents mention or cite which people and references. It is all computed on your machine — and the same connection graph feeds retrieval, so the agent follows links between related documents to find context a plain search would miss. 08 What can it connect to? Three ways. Read-only connectors sync from 20+ sources (cloud storage, databases, Google Workspace, Microsoft 365, Dropbox, Confluence, Notion, a web crawler). MCP 'connected accounts' both read and act in Slack, Teams, Discord, Notion, Jira & Confluence, GitHub, GitLab, Linear and HubSpot. And messaging bridges connect Slack, Telegram, WhatsApp and Instagram. It can also run as an MCP server itself. 09 How is it governed across an organisation? Enterprise and Government editions add a control plane. An admin runs a dashboard for users, groups, policies, knowledge-base access, UI layout and audit. Identity flows through OIDC or SAML SSO with SCIM provisioning. A granular RBAC policy is pushed to every managed client and enforced locally — controlling which models, tools, workflows, connectors, data and even UI surfaces each group may use — with fleet-wide audit. 10 Is AssistantGeneral GDPR compliant? Yes — by architecture. When run locally, all data including prompts, documents, conversation history and configuration stays on your infrastructure. There is no data-processor relationship to manage with Royal Softworks and no DPA required, because no data is sent to us. 11 Is it suitable for air-gapped or government environments? Yes. The Government edition packages the same control plane for sovereign, air-gapped deployment: local-first inference with no runtime cloud dependency, no telemetry, encryption at rest, PIN lock, a full audit trail, and centrally governed policy across the fleet. 12 What editions are available? Four: Demo (free evaluation), Pro (the complete agent for one person, no restrictions), Enterprise (Pro plus the admin control plane — SSO, SCIM, RBAC, audit, fleet management), and Government (the Enterprise control plane packaged for sovereign, air-gapped deployment). 13 How many interface languages does it support? Ten: English, Arabic (with full RTL layout), German, Spanish, French, Italian, Japanese, Russian, Serbian and Chinese. The localisation system is community-extensible. ## Still have a question? Pilot deployments, volume licensing and security reviews go straight to the team that builds the product. [Get in touch](https://royalsoftworks.com/contact/)