More control over which tools are exposed to the conversation agents in Assist

[Ramblurr]

Describe the feature

We need more control over which tools are exposed to the conversation agents, particularly the ability to disable builtin Hass tools.

The number of integrations adding tools and bloating the context is growing every release.

It is well known now that the success of Agentic systems comes down to context engineering. We shouldn't be stuffing so much into the Assist agent's context if it can't be used properly.

Use cases

• Search "HassMediaSearchAndPlay" on the forums and you will only find people talking about how their Assist agent is using that tool instead of their actual desired tool to play media.
• Context management will result in more successful interactions with Assist

Anything else?

No response

[mitrokun]

We're not limited to the Assist option. Any interested user can create an LLM API integration and define any entities, tools, prompts, and so on.

I wanted to be able to use LLM in a limited way without overloading the model with unnecessary context (the built-in commands are sufficient for basic actions), so I created an integration for myself that provides access only to selected scripts.