Coding Horizon

The Local AI Repo Everyone Is Quietly Moving To

Sources for every figure, name, path and quotation the finished picture puts on screen. Checked 24 August 2026.

The repository

Open WebUI star count. 149,780 stars, 21,849 forks, at the time of checking. The narration says “almost one hundred and fifty thousand”, which is what the count supports. Created 6 October 2023.

What the project calls itself. The repository description is “User-friendly AI Interface (Supports Ollama, OpenAI API, …)”. The project site is https://openwebui.com.

Latest release at the time of checking. v0.11.0, published 27 July 2026.

What it connects to

Ollama and OpenAI compatible endpoints. The README states it can “Connect any OpenAI-compatible API alongside local Ollama models”, and names LM Studio, GroqCloud, Mistral, OpenRouter and vLLM as examples of providers that speak that API shape.

Ollama’s default address. “Ollama binds 127.0.0.1 port 11434 by default.” The bind address is changed with the OLLAMA_HOST environment variable. This is the endpoint the connection settings shot fills in.

Retrieval and vector databases

Nine vector databases. The README states retrieval works against “9 vector databases and multiple content-extraction engines”. The nine named are Chroma, PGVector, Qdrant, Milvus, Elasticsearch, OpenSearch, Pinecone, S3Vector and Oracle 23ai. These are the nine the retrieval shot cycles through.

Web search providers. The README names SearXNG, Google PSE, Brave Search, Kagi, Mojeek and Tavily among “dozens of providers”. The three marks on screen are SearXNG, Brave and DuckDuckGo.

Users, models and tools

Roles and groups. The README describes “Granular RBAC & User Groups”, where administrators “define detailed roles, groups, and permissions”. This is what the permission matrix shot draws.

Model presets. The README describes wrapping “any base model with custom instructions, tools, and knowledge”, and lists “Filters, Actions, Pipes, Tools, and Skills” as the extension points.

Speech and images. Speech to text is listed as local Whisper, OpenAI, Deepgram and Azure. Text to speech is listed as Azure, ElevenLabs, OpenAI, Transformers and WebAPI. Image generation is listed as OpenAI DALL·E, Gemini, ComfyUI (local) and AUTOMATIC1111 (local).

Evaluation. The README describes a “built-in arena, A/B testing, and ELO-based leaderboards”, which is what the arena leaderboard shot sorts.

Running it

The install commands and ports on screen. The documented default run is:

docker run -d -p 3000:8080 -v open-webui:/app/backend/data \
  --name open-webui ghcr.io/open-webui/open-webui:main

The bundled Ollama image is ghcr.io/open-webui/open-webui:ollama and the GPU image is ghcr.io/open-webui/open-webui:cuda. Every documented variant maps host port 3000 to container port 8080 and mounts a named volume at /app/backend/data.

The data folder. /app/backend/data is the path the documented volume mount targets, which is why the narration’s point about the data folder mattering is drawn as that mount.

Licensing and branding

What the licence is. Open WebUI ships a modified BSD-3-Clause licence. From v0.6.6, effective 19 April 2025, it adds a branding clause, and the project’s own documentation says v0.6.6 and later “is not an OSI-approved ‘open source’ license”. Everything up to and including v0.6.5 remains under the original BSD-3-Clause with no branding requirement.

The clause itself. Condition 4 of the licence prohibits “altering, removing, obscuring, or replacing any ‘Open WebUI’ branding, including but not limited to the name, logo, or any visual, textual, or symbolic identifiers”, in any deployment or distribution.

The exemptions. The same clause exempts deployments where “the total number of end users … does not exceed fifty (50) within any rolling thirty (30) day period”, licensees with “specific prior written permission from the copyright holder”, and licensees holding an enterprise licence that expressly permits the modification.

These three figures, the fifty users, the thirty day window and the v0.6.6 boundary, are the ones the licence shot puts on screen.

Ollama, for the comparison the video draws

Ollama’s own repository. 179,334 stars at the time of checking. The video does not put this figure on screen; it is here because the video’s argument compares the two projects’ roles and it is worth knowing they are the same order of magnitude.

Not put on screen

Figures the narration states that are not rendered as text or numbers in any shot, so that nothing unsourced appears large: