Open WebUI vs LibreChat vs AnythingLLM: which self-hosted ChatGPT should a small team pick?

Which self-hosted ChatGPT-style UI should a small team pick? For most teams, Open WebUI. It runs as one Docker container, has roles, groups, SSO/LDAP, SCIM and two-factor sign-in built in, and Open WebUI's own docs list internal use "with original branding exactly intact" as free. Pick LibreChat if you need to put your own name on it, because it's MIT-licensed, and accept that it runs as a six-container stack. Pick AnythingLLM if the team mostly chats with documents, wants a desktop app, or would rather pay for hosting (from $50 a month). Checked Nov 3, 2026 on each project's own licence, docs and pricing pages.

Which one should a small team pick?

• You want one tool for 5 to 50 people and don't care whose logo is on it: Open WebUI. It's the only one of the three whose docs list SCIM provisioning (users and groups sync from your directory), and it's a single container to run and back up. You can keep using it as the team grows past 50, as long as you leave the Open WebUI name and logo alone.
• You need your company's name on it, or you resell it to clients: LibreChat. MIT lets you rebrand freely, keeping the copyright notice. It also has the broadest list of providers and the deepest agent features (Skills, subagents, a code interpreter).
• The team mainly wants to ask questions about its own documents: AnythingLLM. Workspaces of documents are its centre of gravity, it has a no-code agent builder, and it's MIT. Note that multi-user mode is Docker-only, and full SSO is listed under its Enterprise plan.
• Nobody wants to run a server at all: AnythingLLM Cloud. It's the only one of the three with published hosted prices: Basic $50 a month, Pro $99 a month (read Nov 3, 2026).
• It's just you, on one laptop: AnythingLLM's desktop app, or the new Open WebUI Desktop app (v0.1.4, released Nov 2, 2026).

How do they compare side by side?

Sources: Open WebUI's features page, LibreChat's README and authentication docs, AnythingLLM's README and pricing page. Release dates and star counts are from GitHub. We compared what each project documents; we haven't run a long-term test of all three side by side.

Is Open WebUI free for a company?

Yes, if you keep its branding. Open WebUI's enterprise page has a two-row table: "Standard Use (Internal use with original branding exactly intact.)" is "Free"; "Enterprise Use (White-labeling, rebranding, modifications to original branding…)" needs an "Enterprise License Required".

The rule behind it is clause 4 of the LICENSE, unchanged when we read it again on Nov 3, 2026. It prohibits "altering, removing, obscuring, or replacing any "Open WebUI" branding", except in "deployments or distributions where the total number of end users (defined as individual natural persons with direct access to the application) does not exceed fifty (50) within any rolling thirty (30) day period", with written permission, or with an enterprise licence. So the 50-user number is a limit on rebranding, not on how many people can use it. The full text and its history are in the Open WebUI licence, explained.

Two more things to know. The new Open WebUI Desktop app is a separate project under AGPL-3.0, per its GitHub repository. And the enterprise page says enterprise licences are "available exclusively to registered entities and organizations", with no public price.

What does LibreChat need to run?

More moving parts than the other two. LibreChat's default docker-compose.yml (read Nov 3, 2026) starts six services: LibreChat itself, an admin panel, MongoDB, Meilisearch for conversation search, a pgvector database and its RAG API for document search. That's fine on one server, but it's six things to update and back up instead of one.

In return you get the widest sign-in options of the three. Its authentication docs point to setup guides for "OAuth2 and OpenID Connect, SAML, LDAP, email, and passkeys". The README lists "Built-in Moderation, and Token spend tools" and a browser-based admin panel to "manage users, groups, roles, and configuration overrides".

One practical tip from the same page: in a single-tenant deployment, "the first account registered gets the ADMIN role, so create it yourself before opening registration to others". Also worth knowing: LibreChat's site carried the banner "LibreChat is joining ClickHouse" on Nov 3, 2026. The repository is still MIT-licensed.

Is AnythingLLM free, or do you pay $50 a month?

Both options exist. The desktop app and the Docker server are free and MIT-licensed. AnythingLLM's pricing page offers "Self-host with Docker for free", and hosted Cloud plans for teams that don't want to run anything: Basic at $50 a month (its card says "Just bring an LLM API Key"), Pro at $99 a month, and Enterprise on request.

For a team, mind the limits in its own docs. The Security and Access page is marked "DOCKER VERSION ONLY", and warns: "Once in multi-user mode, you cannot revert back to single-user mode". Multi-user mode has three roles: Admin, Manager and Default, where Default users "can only send chats to workspaces they are explicitly added to". For logins, the self-hosted version offers a "Simple SSO Passthrough" that generates a temporary login link per user; "SSO, RBAC, and more" appear on the pricing page under Enterprise.

How do users log in: SSO, LDAP, roles?

This is where the three differ most for a company with an identity provider (Google Workspace, Microsoft Entra, Okta):

• Open WebUI: "RBAC: Roles, groups, and per-resource permissions", "SSO/OIDC/LDAP", "SCIM 2.0: Automated user and group provisioning" and "Two-Factor Sign-In (MFA)", per its features page. SCIM means new hires and leavers sync from your directory automatically.
• LibreChat: OAuth2/OIDC, SAML, LDAP, email and passkeys, two-factor authentication that an admin can make mandatory for local and LDAP accounts (its authentication docs), plus per-role and per-group permissions in its admin panel. No SCIM is listed.
• AnythingLLM: three fixed roles and a simple SSO passthrough when self-hosted; full SSO is an Enterprise item.

Keep whichever you pick updated. Open WebUI's v0.13.0 release notes (Nov 3, 2026) say it "contains security fixes" and that "We strongly recommend updating all deployments as soon as possible". The same notes warn that it includes database migrations, so back up first.

Where should the team's server run, and what does the GPU cost?

The UI itself is light; the model behind it is what needs a GPU. Any of the three can sit in front of Ollama on one shared machine, and our Docker guide for local AI covers the Ollama + Open WebUI setup. Check which models fit a card with the VRAM calculator.

If nobody has a spare GPU box, rent one. Our price tracker's reading of Nov 3, 2026 (20:20 UTC): a 24 GB RTX A5000 on Runpod on-demand at $0.27 an hour, which is about $197 a month if it runs around the clock (730 hours), and a 24 GB RTX 3090 on Vast from $0.096 an hour on-demand (median $0.16 across 25 offers). Prices move hourly; see today's prices on /cloud-gpu. A team server that only runs during work hours costs a fraction of the 24/7 figure.

Comparing the web UI route with a desktop app? See Open WebUI vs LM Studio. Still choosing the engine underneath? Ollama vs LM Studio covers that layer, and chatting with your documents locally explains how RAG works in these tools.