Open WebUI vs AnythingLLM for a team’s files

Open WebUI and AnythingLLM both publish their source code and can be run on your own server: AnythingLLM under the MIT license, Open WebUI under its own license, which it says is not an OSI-approved open-source license. [7][8][12] Both put a team’s documents next to an AI model, and both connect to Claude with an Anthropic API key. [1][6][9][11] For a small company the deciding question is usually not the model but the files: where they live, how the AI reads them and who may see which one. Here is how each handles that, and a third option, Claude Code on a company file server, which is what we build — so read that section as ours.

Open WebUI

• Files go into knowledge bases. You create a knowledge base, upload files or mirror a local folder into it with Sync directory, and attach it to a model or reference it in a chat with #. The model can search it or take a whole document word for word (Full Context). [1] A companion tool, oikb, keeps knowledge bases in sync with sources such as a GitHub repo, Confluence, an S3 bucket, Google Drive or SharePoint, can split one source across several knowledge bases, and never bypasses knowledge-base access control. [2]
• Access is set per knowledge base. A knowledge base belongs to the person who created it; private means the owner, plus admins under the default setting. Read or write can be granted to users or groups, and “Public” means every signed-in user. [3] Admins can share any knowledge base; other users can share the ones they can edit if they have the Knowledge Sharing permission, which is off unless turned on. [1][21] Folders inside a knowledge base organize files but don’t restrict access: retrieval spans the whole knowledge base. [1]
• Groups and sign-in. Permissions are additive — a user has the union of what their role and groups grant — and admins can open Preview Access to see every model, knowledge base and tool a user can read. [3][4] Sign-in can use OAuth or OIDC (Google, Microsoft, Okta, Keycloak), with SCIM 2.0 for provisioning. [15]
• Agents that act on files. Open Terminal gives the agent a workspace where it can create files, run code and work with documents. [5] Its built-in multi-user mode is free and separates workspaces, but the docs say it is not a security boundary between users; isolated containers per user come from Terminals, which requires an Open WebUI Enterprise License. [5][16]
• Claude. Add a connection with Anthropic’s OpenAI-compatible endpoint and an API key. Anthropic describes that endpoint as intended for evaluation and testing, not production; Open WebUI says tool calling works in practice, and points to a pipe function for Claude’s native API. [6]
• License. Code up to v0.6.5 stays BSD-3. From v0.6.6 the Open WebUI License adds a branding clause: you keep the Open WebUI branding unless you have 50 or fewer users in a 30-day period, have written permission from Open WebUI, or have an enterprise license. [7]

Fits when many people use many models, your identity provider already holds the groups, and the documents can be split into knowledge bases that each have their own readers.

AnythingLLM

• Desktop, Docker or Cloud. The desktop app is a one-click install for one person, with no multi-user support. The Docker version is the one many people use at the same time, with admin and rule-based access for workspaces and documents. [8] AnythingLLM Cloud is a hosted, multi-user version, each customer on an isolated AWS instance managed by Mintplex Labs, with no built-in local model. [18]
• Files go into workspaces. A document attached in a chat stays in that thread and its full text goes to the model; a document embedded into a workspace is retrieved from, and in multi-user mode it is available to every user who has access to that workspace. [9] Third-party data connectors are available. [8]
• Access is set per workspace. Multi-user mode has three roles. Admin: full access to the entire system. Manager: can view all workspaces and manage all properties except the settings for the LLM, embedder and vector database. Default: can only send chats to workspaces they were explicitly added to, and can’t see or edit any workspaces or system settings. Once you switch to multi-user mode, you can’t switch back. [10] It also supports a simple SSO passthrough. [19]
• An agent in your folders. The File System Agent, off by default, can search folders you allow and read or write files there; on Docker those folders are mounted into the container. [17]
• Claude. Enter an Anthropic API key from the Anthropic Console; all Anthropic models are available. [11]
• License. MIT. [12]

Fits when one person wants a private document assistant on a laptop, or a team’s documents split cleanly into workspaces whose members may all read everything inside.

Claude Code on a company file server

This is what we built Trina AX to do, so here is how it works there. Claude Code (and Codex) run on one company server next to the company’s folders — shared/, groups/, projects/ and personal folders — and people use a chat app in the browser, on desktop or phone. [13]

• The AI opens the files. The engine has no shell or file tools of its own; it reaches files only through the server’s tools, which check the person asking every time. [14] Files a person can’t access stay invisible in Files, AI answers, search, previews, version history and trash. [13]
• Who manages what. Owners and admins have management access to team folders, but not to other people’s personal folders, and chats are visible only to their owner. [13][14][20] Removing someone’s folder access doesn’t erase earlier answers. [13] People join through single-use invitation links with their own username and password. [13]
• Shared ways of working. Team skills are SKILL.md files in folders, and folder rules decide who can use them. [13]
• Claude. The company’s Claude account or each person’s own, or an API key; use only accounts and plans authorized for your use. On new hosted teams, one company connection serves the whole team by default; the site notes that this setting doesn’t grant permission under Anthropic’s or OpenAI’s terms, that sharing one person’s subscription across a team can break them, and that a company API key with a monthly limit per person avoids that question. [13]

What it doesn’t do. Its security page calls it a pre-release with no security certification; development mode has no OS isolation, and a production install adds a dedicated account, a sandbox and a firewall. [14] Server operators, including Trina on a hosted plan, can read stored files and chats. [13] What the AI reads goes to the AI provider you connect; web searches, connected services, speech from files if allowed and a periodic address check also leave the server. [13] It runs on a Linux server or an office Mac (no Codex on a Mac), free to install yourself under AGPL-3.0, or hosted by Trina for 3 to 40 people (larger teams by quote), with a free plan for up to 3 people and paid plans from $50 a month. [13]

Fits when the files already sit in company folders whose access differs folder by folder, and you want the AI to open those files as the person asking, not search an index built per collection.

Side by side

	Open WebUI	AnythingLLM	Trina AX
Where the files live	Knowledge bases: uploads, a synced local folder, or sources synced with oikb [1][2]	Documents attached in a chat (that thread) or embedded in a workspace [9]	Folders on the company server [13]
Who sees which file	Per knowledge base: the owner, users and groups granted read or write, or everyone signed in; admins by default [3]	Per workspace: every member; Admins and Managers see all workspaces [9][10]	Per folder, checked as the person asking; owners and admins manage team folders [13][14][20]
How the AI uses files	Search or full text; an agent workspace with Open Terminal [1][5]	Full text when attached, search when embedded; a File System Agent for allowed folders [9][17]	Claude Code opens them through the server’s tools [14]
Many users	Yes; per-user agent containers need an Enterprise License [5][16]	Docker or Cloud, not Desktop [8][18]	Yes; hosted plans for 3 to 40 people [13]
Sign-in	OAuth / OIDC, SCIM [15]	Simple SSO passthrough [19]	Invitation links with a username and password [13]
Claude	Anthropic’s OpenAI-compatible endpoint, or a pipe function [6]	Anthropic API key [11]	Claude account or API key [13]
License	BSD-3 to v0.6.5; from v0.6.6 its own license with a branding clause [7]	MIT [12]	AGPL-3.0 [13]

How to choose

1. Many people, many models, groups in your identity provider, documents you can split into collections → Open WebUI. If more than 50 people use it in a 30-day period, keep its branding or get an enterprise license.
2. One person on a laptop, or a team whose documents split cleanly into workspaces → AnythingLLM: the desktop app for one person, Docker or Cloud for a team.
3. Company folders whose access differs folder by folder, and an AI that should open them as the person asking → Claude Code on a company file server.

Questions people ask

Is Open WebUI free for a company?

Yes, but from v0.6.6 its license asks you to keep the Open WebUI branding unless you have 50 or fewer users in 30 days, written permission from Open WebUI, or an enterprise license. [7]

Does AnythingLLM support multiple users?

The Docker version and AnythingLLM Cloud do, with Admin, Manager and Default roles; the desktop app does not. [8][10][18]

Can Open WebUI and AnythingLLM use Claude?

Yes, both with an Anthropic API key. Open WebUI uses Anthropic’s OpenAI-compatible endpoint, which Anthropic describes as meant for evaluation and testing, or a pipe function for the native API. [6][11]

Can I limit which files each person sees?

In Open WebUI you share each knowledge base with chosen users or groups; folders inside one knowledge base don’t restrict access. [1][3] In AnythingLLM, everyone who can use a workspace can use its embedded documents. [9] On a company file server like Trina AX, access is set per folder and checked on every file the AI opens. [13][14]

Try it

• Open WebUI: its Knowledge [1] and Groups [3] pages.
• AnythingLLM: its Security and Access page. [10]
• Trina AX: Ask your first question and Set folder permissions. Source: github.com/trina-app/trina-ax. Hosting plans: ax.trina.app/#pricing.

Claude is a trademark of Anthropic. Trina AX is made by Trina Inc. and is not affiliated with Anthropic.