Tools

Tools are how LLMs safely interact with the real world outside their own training. They allow an LLM to call explicit, typed functions (at least with OpenTools) that access external services like Coinbase, Alpaca, or Gmail. Instead of guessing, hallucinating, or relying on stale training data, a model can request fresh, authoritative information and act on it. In OpenTools, tools are first-class citizens. These are structured capabilities ensuring easier setup and configuration into your application by providing structured output, concise error-handling and simpler auth handling under a unified methodology.

---

​

What is a tool?

Simply put, a tool is a function. Unfortunately, tools can look very different across providers and frameworks, as there are few standards governing how they should be defined or exposed to LLMs. OpenTools defines tools around a small set of enforced principles:

• Structured input and output schemas, governed by a single domain (for example trading.schemas)
• Clear, unified error handling via a shared OpenToolsError
• A stable name and description designed explicitly for LLM consumption (ToolSpec)

In OpenTools, tools are not raw API calls and not prompt tricks.
They are explicit, typed capabilities that a model may choose to invoke. From the model’s perspective, a tool is inherently a brief JSON description, but for you it becomes a Pythonic function where capabilities can be added. OpenTools ensures that there is no prompt parsing, no manual JSON handling, and no provider logic leaking into your application code.

---

​

Why tools exist?

LLMs are incredibly useful for useful applications such as answering queries quickly and accurately but only when they have the right context to use. This is when tools can be helpful. They allow you to gain relevant and updated information even when the model’s training date is months before the release of this data! In this case, OpenTools aims to amplify tool use cases by organising them not by provider alone but in application, such as integrating Alpaca and Coinbase into Trading, so that developers can rapidly deploy their projects in minutes. Tools also ensure structured output and execution as they are governed outside the influence of the LLM’s inherently probabilistic nature and using our pre-made tools, it will ensure that only relevant context reaches your preferred model without bloating the information you really want to build on top of.

---

​

Why use our tools?

OpenTools is not a simple wrapper around an API endpoint. It comes with key improvements in the form of error-handling, auth configuration and structured output. This essential difference allows for a developer to easily integrate our toosls:

1. It handles many different Error types which are specified under the OpenToolsError that will showcase what exactly happened in the flow to ensure minimal disruptions whether it be an authentication or model provider issue.
2. Auth is also made painfully simple so that you can just add your API keys and run!
3. Output is unified across domains so that you can expect the same output making it incredibly intuitive and easy to build on top of without having to handle extra edge-cases. We also support a minimal mode so that you can use only the output you need, reducing your token count along the way.

---

​

A quick example

A concrete demonstration is detailed below, where the tool is registered and called allowing for the model to answer a query more precisely. This loop can actually continue several times and not just once in this simple example. The model could, for example, call a tool to determine positions available before executing a trade allowing for a more sophisticated agent which can be done reliably using one of the frameworks OpenTools has integrated with.

---

​

Next steps