Add language model base and work out interactions with planning manager

[collijk]

This PR adds the base abstraction for the language model and adjusts some of the interface of the Planning model to create their communication contract.

[vercel[bot]]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Ignored Deployment

Name	Status	Preview	Comments	Updated (UTC)
docs	⬜️ Ignored (Inspect)			May 5, 2023 3:32pm

[collijk]

Some thoughts on how this will work:

# This may happen "client" application side before we create the Agent, I'm thinking

planner = Planner(...)
language_model = LanguageModel(...)

initial_user_input = ... # Application logic for this
objective_prompt = planner.construct_objective_prompt_from_user_input(initial_user_input)
model_response = language_model.determine_agent_objective(objective_prompt)

agent_objective = model_response.content  # This should be parsed into a standard format already

# This may happen on the "server" side (across the message broker interface)
# ...Update API budget, etc. ...



agent = Agent(
    agent_objective, 
    planner,
    language_model, 
    ...,
)
agent.run()

class Agent:

    def __init__(self, agent_objective, planner, language_model, ...):
        self.objective = agent_objective
        self.planner = planner
        self.language_model = language_model
        ...

    def run():
        while True:
            # Logic to construct the context (checking budget, looking for memories, getting user input)
            planning_prompt_context = PlanningPromptContext(...)
            planning_prompt = self.planner.construct_planning_prompt_from_context(planning_prompt_context)
            action_plan_response = self.language_model.plan_next_axtion(planning_prompt)

            action_plan = action_plan_response.content
            
            # ...Update API budget, check for user confirmation, update memory, etc.
            
            # Logic to construct the context ...
            self_feedback_prompt_context = SelfFeedbackPromptContext(...)
            self_feedback_prompt = self.planner.get_self_feedback_prompt(self_feedback_prompt_context)
            self_feedback_response = self.language_model.get_self_feedback(self_feedback_prompt)
            
            self_feedback = self_feedback_response.content

            # ...Update API budget, resolve action plan with self feedback, update memory, do next action, etc.

[Boostrix]

People interested in learning more about ongoing talks about the planning manager, refer to: #3593