Open WebUI Tools: A Modular Toolkit for Turning Open WebUI into a Workstation

The gap this fills in an Open WebUI deployment

Open WebUI gives you a chat interface over local or remote models. It does not give you arXiv search, YouTube embedding, Pexels media lookup, ComfyUI image and audio generation, or a planner agent that delegates to sub-agents. Those are the things this repository adds. The README describes the collection as a way to "Transform your Open WebUI instance into a powerful AI workstation", and the directory layout backs that up: tools/, functions/ and filters/ each hold standalone Python files that plug into a running instance. The audience is people who already have Open WebUI running and are comfortable pasting Python into a workspace editor. If you have not installed Open WebUI, nothing here applies to you.

Three extension types and what each one can do

Open WebUI distinguishes tools, function pipes and filters, and the repository mirrors that split. Tools are callable capabilities the model can invoke: arXiv Search, Perplexica Search, Pexels Media Search, YouTube Search & Embed, OpenWeatherMap Forecast, and several image and video generators. Function pipes sit in front of the model and change how a conversation is routed or processed: Planner Agent v3, arXiv Research MCTS, Multi Model Conversations v2, Resume Analyzer, Mopidy Music Controller, Letta Agent, Perplexica Pipe, Google Veo, and MiniMax LLM Pipe. Filters intercept messages: Doodle Paint opens a paint canvas before each send, Prompt Enhancer rewrites prompts, Semantic Router picks a model, Full Document handles file processing, Clean Thinking Tags strips artifacts, and OpenRouter WebSearch Citations adds citation handling. The distinction matters because a tool needs the model to decide to call it, while a filter runs whether the model wants it or not. Choosing the wrong type is the most common way to get surprising behaviour.

Installation: the hub path and the copy-paste path

The README gives two routes. The recommended one is the Open WebUI Hub: visit the author's page at openwebui.com/u/Haervwe, click Get on a tool, and follow the prompts inside your instance. The manual route is to copy .py files from tools/, functions/ or filters/, then in Open WebUI go to Workspace > Tools/Functions/Filters, paste the code, give it a name and description, and save. There is no pip install, no requirements.txt step described, and no CLI. Prerequisites are Open WebUI 0.6.0+ and Python 3.8+. Optional dependencies named in the README are ComfyUI for image and music generation, Mopidy for the music controller, and API keys including Hugging Face and Tavily. The configuration section says most tools need minimal setup and lists API keys, ComfyUI integration, model selection and filter setup as the areas to configure, but it does not enumerate which key each tool needs. That gap is the main friction point: you find out what a tool requires by reading its source.

Where the collection is thinner than the list suggests

The README advertises "20+ specialized tools and functions" and the table of contents names 23 entries, but the detailed documentation is truncated in the supplied material, so most sections cannot be verified from what is available. Two claims stand out as worth checking before you rely on them. First, arXiv Search is described as requiring no API key, which is unusual for a search tool and worth confirming against the source. Second, Google Veo is documented as accepting "only one image supported as input" for image-to-video, a hard constraint that rules it out for multi-frame workflows. The broader failure mode is version drift: Open WebUI is a fast-moving project, and a filter that patches message flow can break when the host changes its internal interfaces. Because the repository has no published releases, there is no changelog to tell you which Open WebUI version a given file was written against.

ComfyUI tools versus hosted media APIs

The media tools split into two philosophies. ComfyUI Image-to-Image (Qwen Edit 2509), ComfyUI ACE Step 1.5 Audio, ComfyUI Text-to-Video and Flux Kontext ComfyUI expect you to run ComfyUI yourself. The README notes the text-to-video tool uses a "default WAN 2.2 workflow". You own the GPU, the queue and the workflow JSON, which means you can change the graph, but you also own the failures when a node is missing or a model file is not where the workflow expects it. Atlas Cloud Media Generator and Hugging Face Image Generator take the opposite approach: send a request to a hosted API and get media back. Less setup, no GPU, but you depend on someone else's uptime and pricing. The README lists OutageDeck Provider Status as a tool for checking provider health and incident timelines without an API key, which suggests the author expects provider outages to be a real operational concern.

Maintenance, licence and the cost of pasting code

The repository is MIT licensed, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained. That is the standard permissive position; it is not legal advice, and if you redistribute modified files inside a product you should read the licence text yourself. The maintenance cost is less conventional. Because installation is copy-paste into the Open WebUI workspace rather than a package dependency, upgrades are manual: you re-copy the file and re-enter its configuration. Nothing tracks which version you installed. The repository is not archived and was last pushed in August 2026 according to the metadata, so development appears active, but with no releases there is no semantic version to pin. Budget for reading diffs yourself when you update a tool you depend on.

Who should adopt this, and what to check first

Adopt it if you run Open WebUI 0.6.0+ and want a specific capability now: arXiv search for a research workflow, ComfyUI-driven image or music generation, or the Planner Agent v3 pipe for delegated multi-agent runs. Skip it if you want a supported product with release notes, or if you are not comfortable opening a .py file to find out which environment variables it reads. Before installing, open the file for the tool you want and check three things: the external services it calls, the API keys it expects, and whether it assumes a ComfyUI instance on a particular host and port. The README's configuration section is a starting point, not an inventory, and the detailed per-tool documentation is the place where that inventory would live if it were complete.

Editorial conclusion

Adopt this if you already run Open WebUI 0.6.0+ and want specific capabilities (arXiv search, ComfyUI media generation, a planner agent pipe) without writing integrations yourself. Do not adopt it if you need a supported product with versioned releases: the repository ships no releases, and each tool is pasted into the Open WebUI workspace by hand. Before installing anything, open the individual .py file you intend to use and check which API keys and external services it expects, because the README's configuration section names only Hugging Face and Tavily as examples and does not document the full set.

Official sources

1. Haervwe/open-webui-tools on GitHub
2. Issues
3. License: MIT
4. README