Skip to content
Is there an AI for this?

Field note · Technical

Open WebUI, AnythingLLM, Onyx: which is for what

Three self-hostable projects that all get described as "private ChatGPT" and are built for three different jobs. What each one is actually shaped for, the licence differences that matter, and how to choose without installing all three.

Source
Written by us. 3 of 11 statements are anchored in a fetched document.
Verified
24 Aug 2026
Confidence
Medium

01They are three different shapes

The three projects overlap enough to appear in the same shortlist and differ enough that the shortlist is usually wrong. Open WebUI is a chat interface that grew retrieval. AnythingLLM is a document workspace that grew chat. Onyx is an enterprise search system that answers in chat.

That sentence is the whole comparison. Everything below is the consequence of it.

DimensionOpen WebUIAnythingLLMOnyx
Starts fromA chat interface over any modelA workspace of uploaded documentsConnectors into existing systems
Best whenStaff need a private assistant and models will changeSmall teams need per-project document setsThe corpus already lives in Confluence, Drive, Slack, a ticket system
Retrieval modelPer-user or per-workspace knowledge collectionsWorkspace-scoped document setsIndexed connectors with source permissions
Weakest atLarge heterogeneous corporaEnterprise connector coverageBeing lightweight
Operational weightOne or two containers to startOne container to startA multi-service deployment from the start
What each project is shaped for — our assessment, not a vendor comparison chart

02The licence differences are real

"Open source" is doing a lot of work in these three descriptions, and the differences show up exactly where an enterprise cares: rebranding, redistribution, and which parts of the product are gated.

  • ASSESSMENT

    GitHub reports AnythingLLM’s repository licence as MIT.

    Label withdrawnWritten as a fact resting on a fetched document, and shown as an assessment here because no stored evidence record for it could be resolved when this page was rendered.

  • FACT

    GitHub reports Open WebUI’s repository licence as "Other" rather than a recognised SPDX identifier, which is the signal to read the licence file itself.

  • ASSESSMENT

    Open WebUI’s licence adds a branding condition above fifty users in any rolling thirty-day period.

    Label withdrawnWritten as a fact resting on a fetched document, and shown as an assessment here because no stored evidence record for it could be resolved when this page was rendered.

  • FACT

    GitHub reports Onyx’s repository licence as "Other" rather than a recognised SPDX identifier.

  • ASSESSMENT

    We assess a licence reported as "Other" as a required reading step rather than a red flag: it means the project ships terms that no standard identifier describes, which may be permissive, source-available, or dual.

    BasisRests on the two GitHub licence records cited above and on the Open WebUI branding clause, which is an example of a term a bare SPDX identifier would not have carried.

03What they are written in, and why it matters on a Tuesday

  • FACT

    GitHub reports AnythingLLM’s primary language as JavaScript; Open WebUI’s and Onyx’s as Python.

  • ASSESSMENT

    We assess the primary language as a proxy for who on your team can debug the system at 17:00 on a release day, not as a quality signal.

    BasisRests on the three GitHub language records cited above, read against the skills your team already operates. The comparison is between your team and the codebase, not between the codebases.

  • ASSESSMENT

    AnythingLLM publishes a Docker quick start, so the operational cost of trying it is one container and an afternoon.

    Label withdrawnWritten as a fact resting on a fetched document, and shown as an assessment here because no stored evidence record for it could be resolved when this page was rendered.

04How to choose in one afternoon

  • RECOMMENDATION

    We recommend Open WebUI when the requirement is a private assistant for staff, the corpus is small or absent at first, and you expect to change models.

    BasisDepends on the brief having a chat objective, few connectors and no large existing corpus. Rests on the shape table above and on Open WebUI’s documented OIDC support, which is what makes it survivable at organisation scale.

  • RECOMMENDATION

    We recommend AnythingLLM when small teams need separate document workspaces and nobody will operate a multi-service deployment.

    BasisDepends on low technical capability and a workspace-shaped corpus. Rests on the shape table and on the single-container Docker quick start cited above.

  • RECOMMENDATION

    We recommend Onyx when the documents already live in systems with their own permissions and the real requirement is search across them rather than chat.

    BasisDepends on the brief naming existing systems to connect and on there being an owner for a multi-service deployment. Rests on the shape table above.

  1. Write down the corpus: where the documents are now, and who may read each one.
  2. If the answer names three or more existing systems, start with Onyx.
  3. If it names one shared drive and fewer than five teams, start with AnythingLLM.
  4. If it names nothing yet, start with Open WebUI and revisit once staff have asked a hundred questions.

05Related

Use cases

Stacks

Tools

Deployment reports

More field notes

Ask this as a question

  1. 01Should we use Open WebUI, AnythingLLM or Onyx for our internal AI assistant?

06Evidence

4 records
  1. 01–02
    Tier 4official repository / model cardGitHub

    GitHub API — open-webui/open-webui

    https://api.github.com/repos/open-webui/open-webui

    FreshRetrieved 25 Aug 2026sha256:7be5bffa13e2

    2 records ·

  2. 03–04
    Tier 4official repository / model cardGitHub

    GitHub API — onyx-dot-app/onyx

    https://api.github.com/repos/onyx-dot-app/onyx

    FreshRetrieved 25 Aug 2026sha256:47cebe16baeb

    2 records ·