needs a /provider command

[nairbv]

we should have a /provider command. This will be implemented with something like:

    provider = model["provider_code"]
    if provider == 'anthropic':
        provider_client = AsyncOpenAI(
            api_key=os.getenv("ANTHROPIC_API_KEY"),
            max_retries=0,
            base_url="https://api.anthropic.com/v1/",
        )
        client = OpenAIAgent(provider_client, system_prompt=system_prompt, model=model["model_code"], tools=tools, messages=messages)
        return client
    elif provider == 'openai':
        provider_client = AsyncOpenAI(api_key=os.getenv("OPENAI_API_KEY"), max_retries=0)
        client = OpenAIAgent(provider_client, system_prompt=system_prompt, model=model["model_code"], tools=tools, messages=messages)
        return client
    elif provider == 'deepseek':
        provider_client = AsyncOpenAI(
            api_key=os.getenv("DEEPSEEK_API_KEY"),
            base_url="https://api.deepseek.com",
        )
        client = OpenAIAgent(provider_client, system_prompt=system_prompt, model=model["model_code"], tools=tools, messages=messages)
        return client
    elif provider == 'meta':
        provider_client = AsyncOpenAI(
            api_key=os.getenv("LLAMA_API_KEY"),
            base_url="https://api.llama.com/compat/v1/",
        )
        client = OpenAIAgent(provider_client, system_prompt=system_prompt, model=model["model_code"], tools=tools, messages=messages)
        return client

need to decide if it should preserve or clear message history.

once a new 'provider' is set, the new provider should be used for other 'roles' as well.

This could also be done lazily - set a new default provider and then use it next time a role is changed.

[nairbv]

BTW, we do have support for multiple providers now (e.g. /role anthropic) but adding a /provider command would distinguish the API being used by the agent from the role of the agent. Ideally role should just mean the behavior, system prompt, and set of tools being used by that agent, whereas the provider should be openai/anthropic/deepseek/llama/etc.

[langong347]

I see this problem as an issue where the system prompt needs to be decopuled from agent which also holds an API client:

vibecoder/vibecoder/main.py

Lines 236 to 243 in b8a9e73

	async def switch_role(self, role: str):
	role = role.lower()
	if role == self.agent_type:
	self.print(f"🤖 Already using {role} agent.")
	return
	if role in self.agents_dict:
	if not self.agents_dict[role]:
	self.agents_dict[role] = agents.create_agent_by_role(role)

The expected behavior should be:

• A single client API (openai/anthropic) can support multiple roles which are characterized by the system prompts
• Once an agent is created, we should be able to swap roles without having to recreate the agent

An example use case would be (note: the CLI /provide is a higher level abstraction that can be added later):

from openai import AsyncOpenAI
from vibecoder.agents.agent import OpenAIAgent

client = AsyncOpenAI(api_key="OPENAI_API_KEY")
agent = OpenAIAgent(client, system_prompt=<swe_prompt>, tools=...)

# Later switch role without touching the client
agent.set_system_prompt(<analyst_prompt>)