---
title: "AssistantGeneral — Offline AI for Business"
description: "An AI worker that runs on your own computers, offline: reads your documents, answers from them, builds real spreadsheets and repeats weekly jobs."
url: "https://royalsoftworks.com/products/assistant-general/"
source: "https://royalsoftworks.com"
format: "markdown"
note: "Markdown rendering of the HTML page at `url`. Same content, same canonical URL."
---

[All products](https://royalsoftworks.com/#products)

Private AI automation — offline, on your own computers

# Automate your business. Keep it offline.

AssistantGeneral is an AI worker for your business. It reads your documents, answers from them, builds the spreadsheets and reports you ask for, and repeats the jobs you do every week — on its own. All of it runs on computers you own, with the internet unplugged if you like. Nothing is uploaded. Nothing trains anyone else's AI.

[Request early access](https://royalsoftworks.com/contact/) [See what it can do](https://royalsoftworks.com/products/assistant-general/how-it-works/)

[Free version coming soon — no account, no card](https://royalsoftworks.com/download/)

A question enters a sealed machine on your computer; the agent works inside; a finished document comes out the other side.

01 In plain terms

## You ask for something. It goes and does it.

Not a chat window that gives you advice. You hand it your files, ask in ordinary words, and it comes back with the finished thing — once, or every week without being asked again.

01

“Which of our contracts expire this quarter?”

A straight answer — and the exact page it came from, so you can check it.

answer + sources

02

“Turn these 40 invoices into a Q3 budget.”

A real Excel file, sums already worked out, opened and checked before you get it.

Q3\_budget.xlsx

03

“Summarise this 90-page report for the board.”

One page you can actually hand over, in the tone you asked for.

summary.docx

04

“What did we agree with this supplier last year?”

The clause itself, quoted, from the signed PDF you had forgotten about.

answer + sources

05

“When a customer emails, draft the reply and hold it for my OK.”

It watches the inbox, writes the answer in your words, and waits for you to press send.

runs by itself

06

“Every Monday, gather last week's numbers and write me the summary.”

It does it on its own, week after week, without being asked again.

runs on a schedule

All of it happens on your own computers, with the internet unplugged if you want. Nothing is uploaded, and nobody else ever sees it.

[See everything it does](https://royalsoftworks.com/products/assistant-general/features/)

One question, start to finish

## Watch a request go through.

Seven stages, in order, every single time. Nothing is skipped because the answer already looks plausible — that is the difference between an agent and a chat box.

Seven-stage request pipeline: Route, Plan, Find, Do, Check, Review, Done. Currently showing Route.

Route. Works out which brain to use for this job.

Request classified and dispatched to the configured model — local by default.

1. Route — Works out which brain to use for this job. Request classified and dispatched to the configured model — local by default.
2. Plan — Sketches the steps before touching anything. Planning pass decomposes the request into concrete steps ahead of any tool call.
3. Find — Digs the facts out of your own documents. Hybrid retrieval: vector + BM25 fused with Reciprocal Rank Fusion, reranked, deduplicated.
4. Do — Does the work — builds the file, runs the numbers. Tools execute under the three-tier permission model; risky ones always confirm.
5. Check — Makes sure every claim has a real source behind it. Groundedness gate; unsupported claims trigger a closed re-retrieval pass.
6. Review — Re-reads its own work and fixes what's wrong. Review pass scans for errors, placeholders and inconsistencies, then repairs them.
7. Done — Hands you the answer, and shows its working. Answer plus a full inference trace: every step, tool call and source.

02 Why it's different

## Until now, you had to pick one.

Cloud AI is clever, but every document you hand it is copied to a company you have never met. The private alternatives keep your files safe and can barely read them. If your work is confidential, neither one was a real option.

AssistantGeneral is the missing corner: as capable as the cloud, as private as an unplugged laptop.

Jobs that repeat themselves

## Build it once. It runs every week.

A workflow is a stack of steps. Drag them into the order you want, press play, and watch the job go through. This is the real builder, in miniature — rearrange it and see.

Steps

Every Monday, 8am

Read last week's files

Work out the numbers

Build the spreadsheet

Put it in the shared folder

Drag a step onto the board — or use the arrows on a card to move it without a mouse. Every step is in the job.

1 Every Monday, 8am Starts the job on a schedule, or when a file lands in a folder.

2 Read last week's files Pulls the documents it needs straight off your own drive.

3 Work out the numbers The model does the reasoning, on your hardware.

4 Build the spreadsheet Writes a real .xlsx — formulas, sheets and all.

5 Put it in the shared folder Saves it, emails it, or posts it wherever it belongs.

The real builder has more blocks than this — connectors to twenty-odd systems, branches, approvals — but the shape is the same: steps in an order, running on your own machine, with nothing leaving it.

03 Inside the app

## This is the real thing.

No mockups, no marketing renders. Every screen below is AssistantGeneral running on an ordinary computer — the same build you can try yourself.

Chat & Agent

Visual Workflow Builder

Workflow Results

Agents & Skills

Data Connectors

Messaging Bridges

![AssistantGeneral main chat interface showing sessions sidebar, deep search toggle, and Send to Workflow button](https://royalsoftworks.com/ag/home.png)

Multi-session chat with the agent — toggle knowledge base deep search, send to workflow, or switch to voice mode. LLM, speech, and TTS status live in the footer.

04 The map of your files

## See what your own files know.

Point it at a folder and it draws you a map: your documents sorted into named piles, the subjects that keep coming up, the people and companies that appear together, and which document quotes which rule. Things buried in your own files become something you can simply look at.

The map is not just a picture, either. When you ask a question, it follows those same threads and brings back related documents an ordinary search would miss.

For your IT team

- **Clustering** — On-device k-means over chunk embeddings; clusters auto-labeled by the LLM; linked-cluster edges from centroid cosine similarity
- **Knowledge graph** — Documents joined by shared topics, tags, categories and keywords; retrieval walks the graph to expand results with related sources
- **Views** — Six D3 views — cluster map, topic cloud, timeline, persona network, entity graph, retrieval trace — plus per-source insight chips in chat; all computed locally

Insights · computed locally

Clusters

Topics

Graph

Example cluster map: seven labeled document clusters, sized by content, with links between related clusters.

Vendor contracts 620 chunks renewal liability indemnity SLA click a cluster

An illustration of the Insights overlay. In the app it's live D3 — draggable, zoomable, clickable — computed entirely from your own documents, on your own machine.

Explore AssistantGeneral

## Want more detail?

Everything below is written in the same plain language, in the order most people ask about it.

[Features ### Everything it does Builds real spreadsheets and reports, answers from your own files, works with the apps you already use, and will talk to you out loud. View](https://royalsoftworks.com/products/assistant-general/features/)

[Use cases ### Is it for my line of work? Law, healthcare, finance, factories, government and defence — the same assistant, set up differently for each. View](https://royalsoftworks.com/products/assistant-general/use-cases/)

[How it works ### Why it doesn’t make things up It plans, finds the facts, does the work, then checks itself before answering — six steps you can watch. View](https://royalsoftworks.com/products/assistant-general/how-it-works/)

[Security ### How your data is protected Nothing leaves your machine, nothing is tracked, everything is encrypted — plus the controls an IT team needs. View](https://royalsoftworks.com/products/assistant-general/security/)

[Whitepaper ### The technical paper For your IT and security people: how it is built, what it touches, and what it never touches. View](https://royalsoftworks.com/products/assistant-general/whitepaper/)

[FAQ ### Common questions Does it need internet? Which AI does it use? Is it really private? Answered plainly. View](https://royalsoftworks.com/products/assistant-general/faq/)

[Compare ### How it compares Side by side with 26 other AI tools, including the big cloud names, on the ten things that matter. View](https://royalsoftworks.com/products/assistant-general/compare/)

•Editions

## One product. Four ways to buy it.

Pro is the whole thing, for one person, with nothing held back. Enterprise and Government add what an IT department needs to run it for everybody — and take nothing away.

### Demo

Evaluation

- Hands-on trial
- Preview the admin console
- No commitment

### Pro

Most popular

Individuals & professionals

- The complete agent, no limits
- Local + cloud models
- All tools, voice & integrations
- Encryption, PIN lock, audit

### Enterprise

Companies under IT governance

- Admin console & policies
- SSO (OIDC/SAML) + SCIM
- RBAC on every client
- Fleet audit & KB control

### Government

Public sector & regulated

- Sovereign, air-gapped
- On-prem control plane
- Full audit trail
- Same governance as Enterprise

04 Full feature comparison

## Every feature, every edition.

Pro is the complete agent for one person. Enterprise and Government stack the governance layer on top — SSO, RBAC, fleet audit — without removing a single capability.

| Feature | Free Evaluation | Pro★Individuals & teams | Enterprise IT-governed orgs | Government Public sector & defense |
| --- | --- | --- | --- | --- |
| Core AI & Models |  |  |  |  |
| Local LLM inference (llama.cpp, any GGUF model — fully offline) | Yes | Yes | Yes | Yes |
| Cloud providers: Anthropic, OpenAI, Gemini, xAI, Groq, Mistral, DeepSeek, Together, OpenRouter, custom endpoint | Yes | Yes | Admin-set | Admin-set |
| Multi-machine / multi-GPU model cluster | No | Yes | Admin | Admin |
| Expert personas (Lawyer, Doctor, Cook, Mechanic, Programmer) | No | Yes | Admin | Admin |
| Custom agents & skill packs | No | Yes | Admin | Admin |
| Agent auto-routing | No | Yes | Admin | Admin |
| Agentic loop: plan → retrieve → act → groundedness check → review → finish | Yes | Yes | Yes | Yes |
| Inference trace lens & activity log | No | Yes | Admin | Admin |
| Knowledge Base & RAG |  |  |  |  |
| Local hybrid RAG (vector + keyword, RRF fusion, cross-encoder reranking, MMR) | Yes | Yes | Yes | Yes |
| Number of knowledge bases / collections | 1 (Global) | Unlimited | Admin-set | Admin-set |
| Groundedness checks & cited sources | Yes | Yes | Yes | Yes |
| Ingest: PDF (incl. scanned OCR), DOCX, XLSX, CSV, TXT, MD, HTML, images | Yes | Yes | Yes | Yes |
| YouTube transcript ingestion | No | Yes | Admin | Admin |
| Knowledge graph across documents | Yes | Yes | Yes | Yes |
| Insights visualizations (cluster map, topic cloud, timeline, persona network, entity graph) | No | Yes | Admin | Admin |
| Retrieval tuning panel (HyDE, MMR, BM25 weights, reranker) | No | Yes | Admin | Admin |
| Shared admin-curated knowledge bases | No | No | Yes | Yes |
| Tools & Document Generation |  |  |  |  |
| File read / list | Yes | Yes | Yes | Yes |
| File write / surgical edit | No | Yes | Admin | Admin |
| Shell command execution (confirmation-gated) | No | Yes | No | No |
| Web search & web page browsing | Yes | Yes | Yes | Yes |
| Multi-page web crawler | No | Yes | Admin | Admin |
| Real .xlsx / .docx / .pptx generation via sandboxed Python (Pyodide) | Yes | Yes | Yes | Yes |
| Domain research: PubMed, Europe PMC, ClinicalTrials.gov, US caselaw, retraction screening | Yes | Yes | Yes | Yes |
| HTTP request / API call tool | No | Yes | No | No |
| Workflows & Automation |  |  |  |  |
| Visual drag-and-drop workflow builder | Up to 5 | Unlimited | Admin-set | Admin-set |
| Scheduled workflow runs (cron / interval) | No | Yes | Admin | Admin |
| Event-triggered workflows (email, file watch, …) | No | Yes | Admin | Admin |
| Custom workflow nodes | No | Yes | Admin | Admin |
| Sub-workflow chaining | Yes | Yes | Yes | Yes |
| Workflow run history with inspectable results | Yes | Yes | Yes | Yes |
| Connectivity |  |  |  |  |
| Data connectors (Gmail/IMAP, SharePoint, OneDrive, Google Drive, Dropbox, S3, Azure Blob, GCS, Jira, Confluence, Notion, GitHub, GitLab, HubSpot, DocuSign, Outlook, Google Calendar, OneNote, PostgreSQL, MySQL, BigQuery, Snowflake, Azure DevOps, web crawler, custom — 25+) | No | Yes | Admin-set | Admin-set |
| Messaging bridges: Slack, Telegram, WhatsApp Business, WhatsApp Personal, Instagram | No | Yes | Admin-set | Admin-set |
| MCP client (connect external MCP servers) | Yes | Yes | Admin-set | Admin-set |
| MCP server (expose tools to IDEs / other clients) | No | Yes | Admin | Admin |
| Voice |  |  |  |  |
| Text-to-speech (on-device Piper / Supertonic TTS) | 1 voice (F1) | All voices | Admin-set | Admin-set |
| Speech-to-text (on-device Whisper STT) | No | Yes | Admin | Admin |
| Full voice conversation mode | Partial | Yes | Admin | Admin |
| Security & Privacy |  |  |  |  |
| 100% offline — no cloud required after install | Yes | Yes | Yes | Yes |
| No mandatory telemetry | Yes | Yes | Yes | Yes |
| AES-256-GCM encryption at rest + PIN lock | No | Yes | Yes | Yes |
| Append-only audit log | No | No | Yes | Yes |
| Sandboxed working folder + three-tier tool permission model | Yes | Yes | Yes | Yes |
| Enterprise Governance |  |  |  |  |
| On-premise admin control plane + dashboard | No | No | Yes | Yes |
| Single sign-on: OIDC & SAML 2.0 | No | No | Yes | Yes |
| SCIM automated provisioning & de-provisioning | No | No | Yes | Yes |
| Granular RBAC — per user / per group / per org | No | No | Yes | Yes |
| Policy push to all managed clients (enforced locally, fail-closed) | No | No | Yes | Yes |
| Fleet-wide audit aggregation | No | No | Yes | Yes |
| Deployment |  |  |  |  |
| Windows 10+ / macOS 12+ | Yes | Yes | Yes | Yes |
| 10 interface languages incl. RTL Arabic | Yes | Yes | Yes | Yes |
| Sovereign / air-gapped deployment | No | Self-managed | Self-managed | Purpose-built |
| On-premise control plane server | No | No | Yes | Yes |

Included~Partial Not included Admin = Available; admin decides on/off via the enterprise policy engine Admin-set = Available; admin selects which options are permitted fleet-wide