feat: Add ToolsAgent for models with tool-calling support

[Heinrich-vanNieuwenhuizen]

Description

This pull request introduces a new ToolsAgent that simplifies the integration of tools into agents powered by language models such as Ollama and OpenAI. The agent supports tool-driven workflows, enhancing the flexibility of LangChain agents.

Dependencies

• Added dependencies for langchain_community, langchain_openai, and langchain_ollama.

Testing

• Unit tests for the ToolsAgent have been added, covering various tools and scenarios to ensure robust functionality.
• Example tests demonstrate integration with the CalculatorTool and memory.

@davidmigloz & @Ganeshsivakumar

[davidmigloz]

@Heinrich-vanNieuwenhuizen Let me know if you have any comments on my changes, otherwise I'll merge it!

[Heinrich-vanNieuwenhuizen]

• a3.1, as it's a more powerful model than llama3-groq-tool-use for most use ca

Hi @davidmigloz , do your tests pass with llama3.1 mine only works with llama3-groq-tool-use. I

I had mixed results without "Removing the custom tool-calling prompt, as the providers already handle this. "

[davidmigloz]

do your tests pass with llama3.1 mine only works with llama3-groq-tool-use?

I left llama3-groq-tool-use for the tools_test.dart. Although with llama3.1 all the tests pass except the "Test ToolsAgent with messages memory" test, which is a known limitation of llama3.1:

We recommend using Llama 70B-instruct or Llama 405B-instruct for applications that combine conversation and tool calling. Llama 8B-Instruct can not reliably maintain a conversation alongside tool calling definitions. It can be used for zero-shot tool calling, but tool instructions should be removed for regular conversations between the model and the user.

I had mixed results without the custom tool-calling prompt

If we add that custom prompt, the actual prompt that arrives to the model will contain two tools definitions. The one from our custom prompt plus the tool calling prompt that Ollama adds (or the other providers). So that will probably confuse the model (and add many unnecessary input tokens). The custom prompt should only be necessary for models that don't support tool calling. In that case, the user can provide a custom system prompt:

final agent = ToolsAgent.fromLLMAndTools(
  llm: llm,
  systemChatMessage: <their custom prompt>,
);

[Heinrich-vanNieuwenhuizen]

@Heinrich-vanNieuwenhuizen Let me know if you have any comments on my changes, otherwise I'll merge it! @davidmigloz Okay looks good you can merge

[davidmigloz]

Merged! Thanks again for your contribution @Heinrich-vanNieuwenhuizen 🙌 I'll release it this week.

[davidmigloz]

I've tested it with mistral-nemo, and it's also working pretty well.