Enterprise AI Platform
Coming Soon

AssistantGeneral

Your organisation's knowledge, at every employee's fingertips.

A complete local AI platform for organisations that need AI to work without the data ever leaving their building. Local LLM inference on any GGUF model, a multi-stage RAG knowledge base over your own documents, a visual workflow builder with cron + webhook + file-watch triggers, an agentic tool-calling loop, MCP v1.27 client, four messaging bridges, and 17+ enterprise connectors — all governed by your network perimeter.

v2.0 ALPHA
In late development
GGUF
Any local model
17+
Enterprise connectors
MCP v1.27
Full client + 3 auth schemes
LOCAL LLM INFERENCERAG KNOWLEDGE BASEVOICE I/OWORKFLOW BUILDERMCP PROTOCOLOPENAI COMPATIBLE APIAIR-GAP CAPABLE0 CLOUD CALLSLOCAL LLM INFERENCERAG KNOWLEDGE BASEVOICE I/OWORKFLOW BUILDERMCP PROTOCOLOPENAI COMPATIBLE APIAIR-GAP CAPABLE0 CLOUD CALLS

Available on

  • Windows
    10 / 11 / Server
  • macOS
    12 Monterey+
Core capabilities

What it does.

01
Local LLM inference

Run any open-source model on your own hardware.

Full llama.cpp integration with GGUF model support — Llama, Mistral, Qwen, Phi, Gemma, DeepSeek and the entire open ecosystem. Quantisation from F16 down to Q4. Vision and multimodal supported. A built-in model converter and quantizer means you're never locked to a vendor catalog. The business outcome: you pick the cost/quality trade-off, not a provider's pricing committee.

  • GGUF model support with F16 / Q8 / Q6 / Q5 / Q4 quantisation
  • Vision and multimodal model support (mmproj projections)
  • Direct Anthropic Claude API integration as an optional remote provider
  • Live GPU VRAM and RAM monitor — predict cost before you run it
  • Built-in model converter and quantizer
01
assistant
$ ag chat
› How many vendor contracts expire this quarter?
▸ Querying knowledge base...
▸ 14 contracts match. Drafting summary.
▸ Citing 8 source documents. Done.
02
RAG knowledge base

Multi-stage retrieval that actually finds the right document.

Hybrid retrieval combines semantic vector search, BM25 full-text and fuzzy metadata matching. Each stage is weighted and tunable. Row-Level Security applies at query time so users only retrieve what they're authorised to see. Indexes 15+ document formats. Result: less time hunting for the right file, more time using what's in it.

  • Hybrid retrieval: vector + BM25 + fuzzy metadata, all tunable
  • 15+ source formats including PDF, DOCX, HTML, JSON, CSV, images
  • Row-Level Security enforced on every retrieval
  • Multiple knowledge bases, independently scoped per team or project
  • Reasoning traces show which documents fed each answer
02
Document
Contract_Acme_v3.pdf
Document
Q3_audit_report.docx
Document
Vendor_terms.xlsx
03
Workflow automation

Compose AI pipelines visually, run them on a schedule or a trigger.

A drag-and-drop node editor (powered by Rete.js) for assembling multi-step pipelines. 13 step types, four event triggers — file watcher, webhook, IMAP email, cron — and per-workflow tool and MCP policies. Workflows export/import as portable bundles so you can version and review them like code.

  • 13 step types: LLM, RAG, tool call, sub-workflow, API call, custom function, branching
  • Four triggers: file_watch, webhook, imap_email, cron
  • 4 built-in presets (LLM Only, LLM+RAG, +Tools, +RAG+Tools)
  • Per-workflow tool and MCP policies — granular control
  • Workflow bundles export/import for versioning and audit
03
04
Agentic tool calling

An iterative reasoning loop that plans, calls tools, and refines its answer.

Smart Mode runs an agentic loop: the LLM proposes an action, an optional confirmation gate gives the user (or a policy) the chance to intervene, the tool executes, output is appended to context, and the loop continues until the model produces a final answer. Configurable max iterations, full reasoning trace, every tool call audit-logged.

  • Multi-turn agentic loop with confirmation gates
  • Smart Mode: iterative LLM refinement up to a configurable cap
  • Every tool call audit-logged with inputs and outputs
  • Reasoning/thinking traces shown as collapsible blocks in the UI
04
05
MCP support

Connect the growing ecosystem of MCP-compatible tools.

Full Model Context Protocol v1.27 client. Discover tools served by any MCP-compliant server, authenticate via OAuth2 PKCE, bearer token or API key, and invoke them from chat or workflows. Three transport modes: stdio, SSE and streamable HTTP. Credential storage is encrypted at rest.

  • MCP v1.27 full client implementation
  • Three auth schemes: OAuth2 PKCE, bearer token, API key
  • Three transports: stdio, SSE, streamable HTTP
  • Automatic tool discovery and invocation across all connected servers
05
github
fs
shell
browser
calc
db
06
Enterprise connectors

Ingest from the tools your business already runs on.

Seventeen-plus connectors out of the box. Microsoft 365 (Outlook, Teams, OneNote, SharePoint), Google Workspace (Drive, Gmail, Calendar, Contacts), Slack, HubSpot, Confluence, Notion, GitHub, Postgres, AWS S3, Azure Blob and a web crawler. OAuth2 PKCE handled natively. DocuSign on the roadmap. Plus any MCP-compatible source.

  • Microsoft 365: Outlook, Teams, OneNote, SharePoint
  • Google Workspace: Drive, Gmail, Calendar, Contacts
  • Business platforms: Slack, HubSpot, Confluence, Notion
  • Code and storage: GitHub, AWS S3, Azure Blob, web crawler
  • Databases: Postgres (more coming)
  • Plus any MCP-compatible server
06
More capabilities

Everything else it does.

OpenAI-compatible local API

Anything that speaks the OpenAI API — Cursor, Continue, internal apps, CI pipelines — works against localhost without modification. No keys, no rate limits.

Four messaging bridges

Telegram (long-polling), WhatsApp Business (Meta Cloud API), WhatsApp Personal (Baileys with QR auth), and Instagram (DMs + story replies + scheduled posts).

Admin/Client topology

One Admin node indexes the org KB and broadcasts an encrypted WebSocket server. Client nodes connect across the fleet — RLS enforced at every query.

Piper TTS, 10 voices

Ten high-quality offline voices (5 male, 5 female). Useful for accessibility, dictation, and voice-driven workflows.

Spreadsheets and documents

Built-in spreadsheet and rich document editors with AI authoring. Export to XLSX, CSV, DOCX or plain text. Powered by ag-Grid.

Scheduled workflows

Cron-style scheduling with full run history: event ID, fired-at, duration, success, error, response. Useful for nightly reports and timed audits.

10 interface languages

English, Arabic (full RTL), German, Spanish, French, Italian, Japanese, Russian, Serbian, Chinese — extensible.

PIN lock and audit log

Machine-bound licensing, optional PIN lock, and a tamper-evident audit log of every prompt, retrieval, and tool invocation.

Event-driven triggers

Workflows can fire from file changes, webhooks (HMAC-verified), incoming IMAP email, or cron — closing the loop on real-world automation.

Licensing

Flexible per-seat pricing for teams of any size.

  • Solo, team and enterprise plans
  • Volume discounts above 25 seats
  • Air-gapped activation for restricted environments
  • No usage-based cloud fees — only what you signed for
Frequently asked

Questions we hear often.

No. After installation everything runs locally — LLM inference, the knowledge base, workflows, voice synthesis and the local API. It is designed for air-gapped and network-restricted environments.

Talk to the team that actually builds the software.

Pilot deployments, volume licensing, product demos, security questionnaires — all handled by engineers and product leads, not a routing layer. We respond within one business day.

Schedule a discovery call
Half-hour walkthrough with someone who built the product — no sales script.
Run a pilot deployment
Full-feature evaluation with guided install, configured for your environment.
Email us directly
sales@royalsoftworks.com — we respond within one business day.

Send us a message

Leave your details and we'll follow up within one business day.