add auto build agent feature #5670
Conversation
| enhanced_sys_msg = GROUP_CHAT_DESCRIPTION.format( | ||
| name=agent_name, members=member_name, code_executor_desc=code_executor_desc, sys_msg=system_message | ||
| ) | ||
| await agent.update_system_message(enhanced_sys_msg) |
There was a problem hiding this comment.
Rather than updating, can we create the agent once all the parameters are ready.
There was a problem hiding this comment.
you are right, l will fix it
|
|
||
| return self._build_agents(use_oai_assistant, code_executor=code_executor, **kwargs) | ||
|
|
||
| async def _build_agents( |
There was a problem hiding this comment.
AssistantAgent has built-in functions for serialization and loading. See example: https://microsoft.github.io/autogen/dev/user-guide/agentchat-user-guide/serialize-components.html#agent-example
I think we should be using the built-in API for agent library.
There was a problem hiding this comment.
you are right, but here I'm not just managing agent configurations, but focus on task configurations. and The base_url and api_key in the agent configuration should not be saved but should be loaded in real-time.
l need save the task config like:{
"building_task": "... ",
"agent_configs": [
{
"name": "...",
"system_message": "...",
"description": "...."
},
....
],
"coding": true,
"code_execution_config": {
"last_n_messages": 2,
"work_dir": "....",
"timeout": 60,
"use_docker": "python:3"
}
}
There was a problem hiding this comment.
Make sense, though consider use a BaseModel for your configuration type so we are type checking and validation automatically.
There was a problem hiding this comment.
I think we should be able to do this without changing AssistantAgent.
When registered tools need to be updated, why not create new agent?
There was a problem hiding this comment.
I mean even if you create a new agent with some tools, the registered tools are hardcoded. However, in the Autobuild class, tools can be dynamically registered to an agent.
There was a problem hiding this comment.
Let's put it inside autogen_ext.auto_build as a separate module.
| self.agent_configs: List[Dict] = [] | ||
| self.open_ports: List[str] = [] | ||
| self.agent_procs: Dict[str, Tuple[sp.Popen, str]] = {} | ||
| self.agent_procs_assign: Dict[str, Tuple[AssistantAgent, str]] = {} |
There was a problem hiding this comment.
Is this for running agents in separate processes?
There was a problem hiding this comment.
no，This code is hard to understand. I will optimize it.
There was a problem hiding this comment.
In case you are interested in concurrent running agents, the Core API is a good way to look into it.
|
I will add test code later. |
There was a problem hiding this comment.
@victordibia would you like to take a look at this? Perhaps it can be a feature in AGS.
| self.agent_configs: List[Dict] = [] | ||
| self.open_ports: List[str] = [] | ||
| self.agent_procs: Dict[str, Tuple[sp.Popen, str]] = {} | ||
| self.agent_procs_assign: Dict[str, Tuple[AssistantAgent, str]] = {} |
There was a problem hiding this comment.
In case you are interested in concurrent running agents, the Core API is a good way to look into it.
Why are these changes needed?
there is an AutoBuild feature on the 0.2 branch( https://microsoft.github.io/autogen/0.2/blog/2023/11/26/Agent-AutoBuild/), and I want to add this feature
Related issue number
#5573
Checks