Docs: improve custom tool guide (#176)

* DocsL improve custom tool guide

* Update tool_guide.md

docs/custom_tool_guides/tool_guide.md

@@ -34,7 +34,7 @@ If you need to install a new library to run your tool, execute the following com
 ```bash
 poetry add <MODULE> --group community
 ```
-### Implementing a Tool
+### Implementing a Langchain Tool
 
 Add the implementation inside a tool class that inherits from `BaseTool`. This class will need to implement the `call()` method, which should return a list of dictionary results.
 
@@ -67,6 +67,84 @@ class ArxivRetriever(BaseTool):
         return [{"text": result}] # <- Return list of results, in this case there is only one
 ```
 
+### Implementing a custom Tool
+
+If you are implementing a custom tool, you can follow the same structure as the Langchain tools. This class will also need to implement the `call()` method, which should return a list of dictionary results with a "text" or "result" field.
+
+#### Calculator Tool
+For example, let's look at the calculator tool:
+
+```python
+from typing import Any, Dict, List
+
+from py_expression_eval import Parser
+
+from backend.tools.base import BaseTool
+
+
+class Calculator(BaseTool):
+    """
+    Function Tool that evaluates mathematical expressions.
+    """
+
+    @classmethod
+    def is_available(cls) -> bool:
+        return True
+
+    def call(self, parameters: dict, **kwargs: Any) -> List[Dict[str, Any]]:
+        math_parser = Parser()
+        to_evaluate = parameters.get("code", "").replace("pi", "PI").replace("e", "E")
+
+        result = []
+        try:
+            result = {"result": math_parser.parse(to_evaluate).evaluate({})}
+        except Exception:
+            result = {"result": "Parsing error - syntax not allowed."}
+        return result
+```
+
+The `call()` method receives a dictionary of parameters, which is defined in the `Parameter_definitions` field in the tool configuration (described in the section below). The parameters are generated by the model and passed to the tool, so the tool should assume that the parameters are in the correct format and handle them. 
+In this case, the calculator tool expects a `code` parameter that contains the mathematical expression to evaluate. We then parse the expression and evaluate it, returning the result.
+
+#### Python Interpreter Tool
+Another example is the Python Interpreter tool:
+
+```python
+import json
+import os
+from typing import Any, Dict, Mapping
+
+import requests
+from langchain_core.tools import Tool as LangchainTool
+from pydantic.v1 import BaseModel, Field
+
+from backend.tools.base import BaseTool
+
+class PythonInterpreter(BaseTool):
+    """
+    This class calls arbitrary code against a Python interpreter.
+    It requires a URL at which the interpreter lives
+    """
+
+    @classmethod
+    def is_available(cls) -> bool:
+        return cls.interpreter_url is not None
+
+    def call(self, parameters: dict, **kwargs: Any):
+        if not self.interpreter_url:
+            raise Exception("Python Interpreter tool called while URL not set")
+
+        code = parameters.get("code", "")
+        res = requests.post(self.interpreter_url, json={"code": code})
+        clean_res = self._clean_response(res.json())
+
+        return clean_res
+```
+
+The python interpreter tool expects a `code` parameter that contains the python code to execute. It then sends a POST request to the interpreter URL with the code and returns the result.
+Note that this tool requires a secret `interpreter_url` to be set in the environment variables, and the same can be done for other tools that require secrets.
+
+
 ## Step 4: Making Your Tool Available
 
 To make your tool available, add its definition to the community tools [config.py](https://github.com/cohere-ai/cohere-toolkit/blob/main/src/community/config/tools.py).