nniback / core /agents /agent_setup.py
Kyrlos11's picture
Upload 842 files
22087d0 verified
Raw History Blame Contribute Delete
10.1 kB
"""
Agent setup from natural language description.
"""
import json
from typing import Optional
from fastapi import APIRouter, Depends, HTTPException
from pydantic import BaseModel
from core.utils.logger import logger
from core.services.llm import make_llm_api_call
from core.utils.icon_generator import generate_icon_and_colors
from core.utils.auth_utils import verify_and_get_user_id_from_jwt
from core.versioning.version_service import get_version_service as _get_version_service
from core.utils.core_tools_helper import ensure_core_tools_enabled
from core.config.config_helper import _get_default_agentpress_tools
from core.ai_models import model_manager
from core.services.supabase import DBConnection
db = DBConnection()
from core.api_models import AgentResponse
router = APIRouter(tags=["agents"])
class AgentSetupFromChatRequest(BaseModel):
description: str
class AgentSetupFromChatResponse(BaseModel):
agent_id: str
name: str
system_prompt: str
icon_name: str
icon_color: str
icon_background: str
async def generate_agent_name_and_prompt(description: str) -> dict:
"""
Generate agent name and system prompt from description.
Args:
description: User's natural language description
Returns:
Dict with keys: name, system_prompt
"""
try:
model_name = "openai/gpt-5-nano-2025-08-07"
system_prompt = """You are an AI worker configuration expert. Generate a name and system prompt for an AI worker.
Respond with JSON:
{"name": "Worker Name (2-4 words)", "system_prompt": "Detailed instructions for the worker's role and behavior"}
Example:
{"name": "Research Assistant", "system_prompt": "Act as an expert research assistant. Help users find and analyze information. Always verify facts and cite sources clearly."}"""
user_message = f"Generate name and system prompt for:\n\n{description}"
messages = [{"role": "system", "content": system_prompt}, {"role": "user", "content": user_message}]
logger.debug(f"Calling LLM for name/prompt generation")
response = await make_llm_api_call(
messages=messages,
model_name=model_name,
max_tokens=2000,
temperature=0.7,
response_format={"type": "json_object"},
stream=False
)
if response and response.get('choices') and response['choices'][0].get('message'):
raw_content = response['choices'][0]['message'].get('content', '').strip()
try:
parsed = json.loads(raw_content)
if 'name' in parsed and 'system_prompt' in parsed:
return {
"name": parsed['name'].strip(),
"system_prompt": parsed['system_prompt'].strip()
}
except json.JSONDecodeError as e:
logger.warning(f"Failed to parse name/prompt response: {e}")
return {
"name": "Custom Assistant",
"system_prompt": f"Act as a helpful AI assistant. {description}"
}
except Exception as e:
logger.error(f"Error generating name/prompt: {str(e)}")
return {
"name": "Custom Assistant",
"system_prompt": f"Act as a helpful AI assistant. {description}"
}
async def generate_agent_config_from_description(description: str) -> dict:
"""
Use LLM to generate agent config (name, prompt, icon, colors) from description.
Runs name/prompt and icon/color generation in parallel for speed.
Args:
description: User's natural language description of what the agent should do
Returns:
Dict with keys: name, system_prompt, icon_name, icon_color, icon_background
"""
logger.debug(f"Generating agent config from description: {description[:100]}...")
try:
from core.utils.icon_generator import generate_icon_and_colors
import asyncio
# Run both LLM calls in parallel
name_prompt_task = generate_agent_name_and_prompt(description)
icon_task = generate_icon_and_colors(name="", description=description)
# Wait for both to complete
name_prompt_result, icon_result = await asyncio.gather(
name_prompt_task,
icon_task,
return_exceptions=True
)
# Handle errors
if isinstance(name_prompt_result, Exception):
logger.error(f"Error in name/prompt generation: {name_prompt_result}")
name_prompt_result = {
"name": "Custom Assistant",
"system_prompt": f"Act as a helpful AI assistant. {description}"
}
if isinstance(icon_result, Exception):
logger.error(f"Error in icon generation: {icon_result}")
icon_result = {
"icon_name": "bot",
"icon_color": "#FFFFFF",
"icon_background": "#6366F1"
}
result = {
"name": name_prompt_result.get("name", "Custom Assistant"),
"system_prompt": name_prompt_result.get("system_prompt", f"Act as a helpful AI assistant. {description}"),
"icon_name": icon_result.get("icon_name", "bot"),
"icon_color": icon_result.get("icon_color", "#FFFFFF"),
"icon_background": icon_result.get("icon_background", "#6366F1")
}
logger.debug(f"Generated config: name='{result['name']}', icon={result['icon_name']}")
return result
except Exception as e:
logger.error(f"Error generating agent config: {str(e)}")
return {
"name": "Custom Assistant",
"system_prompt": f"Act as a helpful AI assistant. {description}",
"icon_name": "bot",
"icon_color": "#FFFFFF",
"icon_background": "#6366F1"
}
@router.post("/agents/setup-from-chat", response_model=AgentSetupFromChatResponse, summary="Setup Agent from Chat Description", operation_id="setup_agent_from_chat")
async def setup_agent_from_chat(
request: AgentSetupFromChatRequest,
user_id: str = Depends(verify_and_get_user_id_from_jwt)
):
logger.info(f"Setting up worker from chat for user {user_id}")
if not request.description.strip():
raise HTTPException(status_code=400, detail="Description cannot be empty")
client = await db.client
from core.utils.limits_checker import check_agent_count_limit
limit_check = await check_agent_count_limit(user_id)
if not limit_check['can_create']:
error_detail = {
"message": f"Maximum of {limit_check['limit']} workers allowed for your current plan. You have {limit_check['current_count']} agents.",
"current_count": limit_check['current_count'],
"limit": limit_check['limit'],
"tier_name": limit_check['tier_name'],
"error_code": "AGENT_LIMIT_EXCEEDED"
}
logger.warning(f"Agent limit exceeded for account {user_id}")
raise HTTPException(status_code=402, detail=error_detail)
try:
config = await generate_agent_config_from_description(request.description)
agent_name = config['name']
system_prompt = config['system_prompt']
insert_data = {
"account_id": user_id,
"name": agent_name,
"icon_name": config["icon_name"],
"icon_color": config["icon_color"],
"icon_background": config["icon_background"],
"is_default": False,
"version_count": 1
}
from core.agents import repo as agents_repo
agent = await agents_repo.create_agent(
account_id=user_id,
name=agent_name,
icon_name=config["icon_name"],
icon_color=config["icon_color"],
icon_background=config["icon_background"],
is_default=False
)
if not agent:
raise HTTPException(status_code=500, detail="Failed to create worker")
agent_id = agent['agent_id']
try:
version_service = await _get_version_service()
agentpress_tools = ensure_core_tools_enabled(_get_default_agentpress_tools())
default_model = await model_manager.get_default_model_for_user(client, user_id)
version = await version_service.create_version(
agent_id=agent_id,
user_id=user_id,
system_prompt=system_prompt,
model=default_model,
configured_mcps=[],
custom_mcps=[],
agentpress_tools=agentpress_tools,
version_name="v1",
change_description="Initial version - created from chat"
)
except Exception as e:
logger.error(f"Error creating initial version: {str(e)}")
await agents_repo.delete_agent(agent_id, user_id)
raise HTTPException(status_code=500, detail="Failed to create initial version")
await agents_repo.update_agent_fields(
agent_id, user_id,
current_version_id=version.version_id,
version_count=1
)
from core.utils.cache import Cache
await Cache.invalidate(f"agent_count_limit:{user_id}")
logger.info(f"Successfully created agent '{agent_name}' (ID: {agent_id}) from chat description")
return AgentSetupFromChatResponse(
agent_id=agent_id,
name=agent_name,
system_prompt=system_prompt,
icon_name=config["icon_name"],
icon_color=config["icon_color"],
icon_background=config["icon_background"]
)
except HTTPException:
raise
except Exception as e:
logger.error(f"Error setting up agent from chat: {str(e)}")
raise HTTPException(status_code=500, detail=f"Failed to setup agent: {str(e)}")