Extending Models with Tools & Skills

Tools let a model search the web, run code, query a database, or look up papers on arXiv during a chat. Skills give it access to specialized procedural instructions without requiring you to put those instructions in the system prompt.

Tools

Types of Tools

Built-in capabilities. Web Search, Image Generation, URL fetching, and Memory are available when configured and enabled by administrators and permitted for your account and model.
Workspace tools. Python scripts registered in the platform. These run on the server when the model calls them.
MCP (Model Context Protocol). Anthropic’s open standard for connecting AI models to external data sources and services. MCP servers expose structured capabilities that models discover and invoke automatically.
OpenAPI servers. Any web service with an OpenAPI specification can be connected as a tool provider.

Enabling Tools Per-Chat

Open a chat
Open Integrations beside the plus button in the message box
Browse or search available tools
Toggle on the tools you want for this session
- Tools enabled this way apply only to the current chat

Enabling Tools Per-Model

Request Workspace access by emailing the CUNY AI Lab team, then go to Workspace > Models once access is enabled.
Edit the model you want to configure
Scroll to the Tools section
Select the default tools for this model
Click Save & Update
- Share each attached tool with the intended users or course group as well as sharing the model.

Community Tool Library

The Open WebUI community library lists tools administrators can install. To request one for the Sandbox, send its link to the CUNY AI Lab team. Examples include

arXiv Search — query academic papers directly from chat. No API key required.
Perplexica Search — web search with inline citations.
Pexels Media Search — find stock photos and videos for presentations.
YouTube Search & Embed — locate and embed instructional videos.

Tip. Browse the community library before building your own tool. Someone may have already solved your problem.

Function Calling

Use a model that supports tool calling and check its Function Calling setting under Advanced Parameters. Native mode lets the model request a tool with structured arguments. Test the chosen model with each tool before sharing the configuration.

Skills

Skills are reusable Markdown instructions for tasks or procedures. With Workspace access enabled, create a skill in Workspace > Skills or choose one shared with you, then attach it to a model using the steps below.

Binding Skills

Go to Workspace > Models
Edit the model
Scroll to the Skills section
Select the skills you want to bind
Click Save & Update

Share attached skills with the intended users or course group as well as sharing the model. For skills attached to a model, set Function Calling to Native in the model’s Advanced Parameters so it can load the instructions. Test a request that uses the skill before sharing the model.

Example Skills for CUNY

Quantitative Methods. Statistical analysis, research design, SPSS and R guidance
Academic Writing. Scholarly conventions, citation practices, genre awareness
Research Ethics. IRB compliance, data privacy, informed consent protocols
Digital Humanities. Text analysis, corpus methods, visualization techniques

Advanced Settings

View details

Building Custom Tools

If the community library does not have what you need, administrators can write custom tools. Each tool requires a name, description, and Python function body. The function runs on the server when the model invokes it.

Open Workspace > Tools and choose Create to get started, if your account has permission. Consult with the CUNY AI Lab team if you are unsure about security implications.

Tool Security

Tools are Python scripts that execute on the server. A poorly written or malicious tool can access system resources, exfiltrate data, or disrupt service. Only install tools from trusted sources. Contact the CUNY AI Lab team before adding community tools to the instance.

Callout

Security reminder. Review any community tool's code before installing it. Tools run with server-level access. If you are not comfortable evaluating Python code, ask the CUNY AI Lab team to review it for you.

Additional Resources

Open WebUI Tools Documentation — official reference for tool development and configuration
Model Context Protocol (MCP) — Anthropic’s specification for connecting AI models to external data
Open WebUI Community Tools — browse and download pre-built tools

← Return to Knowledge Bases | Continue to Roles & Permissions →