--- title: "Against self-hosted AI servers · AssistantGeneral" description: "The office's own ChatGPT, if someone runs it." url: "https://royalsoftworks.com/products/assistant-general/compare/self-hosted/" source: "https://royalsoftworks.com" format: "markdown" note: "Markdown rendering of the HTML page at `url`. Same content, same canonical URL." --- Competitive comparison · 3 of 9 # Against self-hosted AI servers. The office's own ChatGPT, if someone runs it. | Product | Own hardware | Your files | Workflows | Connectors | Messaging | MCP | Voice | Admin (theirs) | Admin (yours) | No internet | Desktop | Pricing | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | AssistantGeneral★ This product · checked against our code, 12 October 2026 Private beta, Windows only. The admin console runs on one of your own machines; company sign-in (SSO, SCIM) only in pilots. Works with no internet once set up — activating a licence needs a connection once. | Yes | Yes | Yes | Yes | Yes | Yes | Yes | No | Yes | Partial | Yes | Private beta — no published price yet | | Local AI — self-hosted servers | | | | | | | | | | | | | | Open WebUI Sign-in via SSO, LDAP and SCIM, with roles. Often sold preinstalled on a GPU box. Audit log not confirmed; desktop app in early alpha. | Yes | Yes | No | Partial | No | Yes | Yes | No | Yes | Yes | Partial | Free (branding locked above 50 users) / Enterprise licence | | Zylon (PrivateGPT) A private AI server: needs 80–96 GB of graphics memory. | Yes | Yes | Partial | Yes | No | Yes | Not checked | No | Yes | Yes | No | PrivateGPT free (Apache-2.0) / Zylon by quote, fixed cost | | LocalAI API keys, roles and quotas; no single sign-on or audit log found. Messaging via LocalAGI. | Yes | Yes | Partial | Yes | Yes | Yes | Yes | No | Partial | Yes | No | Free / Managed | | Ollama Native Windows app since July 2025; its team controls cover Ollama's cloud models only. | Yes | No | No | No | No | No | No | Partial | No | Yes | Yes | Free / Pro $20 / Max $100 / Team $500/mo | | SillyTavern A hobbyist front end. | Yes | Partial | No | No | No | No | Yes | No | No | Yes | No | Free | Yes~Partly, or only in some editions No?Not checked — unknown, not no - **Own hardware:** — The AI model can run on hardware you own — your PC or your own server. - **Your files:** — Searches your own documents and answers from them. - **Workflows:** — Multi-step automations; a visual builder counts as full. - **Connectors:** — Ten or more built-in data connectors. - **Messaging:** — Works as a bot in Slack, Teams, WhatsApp or Telegram. - **MCP:** — Model Context Protocol: plugs into outside tools. - **Voice:** — Speech in and out (partial: cloud-only or one direction). - **Admin (theirs):** — Admin console run by the vendor, in its own cloud: sign-in, roles, audit. - **Admin (yours):** — Admin console and policy you host on your own machines. - **No internet:** — Works with no internet connection at all. - **Desktop:** — Native desktop application. Rows checked on 30 September 2026 against each vendor’s own pages (sources at the end of this page); our own row against our own code on 12 October 2026. Off the table since the last version: Flowise reached end of life on 31 August 2026. Backyard AI's desktop app is discontinued. Enchanted was superseded by another app from the same developer. Prices and features change quickly — check before you buy, and tell us if a cell is wrong. Contents 1. 01Self-hosted servers: the office’s own ChatGPT, if someone runs it 01 ## Self-hosted servers: the office’s own ChatGPT, if someone runs it Open WebUI, LocalAI and Zylon put a private assistant on a server you control. Open WebUI is the one to beat: free for up to 50 users, with LDAP, single sign-on, SCIM, roles, voice and MCP — and integrators now sell it preinstalled on a GPU box, which removes most of the setup. What remains is that somebody has to own that server — its updates, its backups, its graphics card — and staff use it through a browser. AssistantGeneral’s firm-wide governance also runs on a machine you host. The difference is that it is the same desktop app with its admin console switched on, not a separate system to stand up. - Mature identity and access control, often free - Needs a server and someone responsible for it; staff work in a browser - Appliances of the Zylon kind need 80–96 GB of graphics memory [Previous Against desktop AI apps.](https://royalsoftworks.com/products/assistant-general/compare/desktop-apps/) [Next Against EU workspaces and company platforms.](https://royalsoftworks.com/products/assistant-general/compare/company-platforms/) [All of AssistantGeneral compare](https://royalsoftworks.com/products/assistant-general/compare/)