AI Toolkit

Chat, tools and spoken conversations for your own scripts, in a few lines of Berry. The module knows no services, models or prompts: your script brings all of that, so it fits any service that speaks the OpenAI format, online or on your own computer.

import ai

var chat = ai.chat({'url': url, 'model': model, 'key': '{secret:openai_key}', 'system': 'Answer briefly.'})
chat.tool('get_time', 'Tells the time', {}, / args, done -> done({'hour': hour(), 'minute': minute()}))
chat.ask('What time is it?', / text, status -> self.show(text, status))

chat keeps the conversation, runs the tools the model asks for, and hands you the final answer. chat.cancel() drops an open question and chat.forget() starts over. A failed request gives nil and a status, and chat.detail holds the service's own message.

On a TC002, ai.converse runs the whole spoken loop and tells you what to draw:

ai.converse(chat, {'hear': {'url': stt_url, 'language': 'en'}, 'followup': true, 'wait': 5},
            / state, talk -> self.changed(state, talk))

state is listening, thinking, speaking, done or error. On error, talk.failed names the step and talk.status the reason. ai.hear and ai.say do one step on their own.

To use it, add this line to your script's header, with the twelve characters at the end of this page's address as the ID. Installing your script then fetches the module too:

# @requires ai <id>

Needs AWTRIX NG 1.2.3 or newer. It takes about 23 KB of memory, which is a lot on a TC001. Lumen Assistant is a complete app built on it.