Berack/upo-app-agents
0
0
Fork 0
Code Issues Pull Requests 3 Actions Packages Projects Releases Wiki Activity

Merge branch 'main' into 21-team-monitoring

Browse Source
This commit is contained in:
Giacomo Bertolazzi
2025-10-12 23:41:19 +02:00
parent c7569df06e 862525cc62
commit 8607814f61
62 changed files with 785 additions and 609 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/app/agents/__init__.py
View File

@@ -1,6 +1,5 @@
from app.agents.models import AppModels
from app.agents.predictor import PredictorInput, PredictorOutput, PredictorStyle, PREDICTOR_INSTRUCTIONS
from app.agents.predictor import PredictorInput, PredictorOutput
from app.agents.team import create_team_with
from app.agents.pipeline import Pipeline
__all__ = ["AppModels", "PredictorInput", "PredictorOutput", "PredictorStyle", "PREDICTOR_INSTRUCTIONS", "create_team_with", "Pipeline"]
__all__ = ["PredictorInput", "PredictorOutput", "create_team_with", "Pipeline"]

107
src/app/agents/models.py
View File

@@ -1,107 +0,0 @@
import os
import ollama
from enum import Enum
from agno.agent import Agent
from agno.models.base import Model
from agno.models.google import Gemini
from agno.models.ollama import Ollama
from agno.tools import Toolkit
from agno.utils.log import log_warning #type: ignore
from pydantic import BaseModel
class AppModels(Enum):
"""
Enum per i modelli supportati.
Aggiungere nuovi modelli qui se necessario.
Per quanto riguarda Ollama, i modelli dovranno essere scaricati e installati
localmente seguendo le istruzioni di https://ollama.com/docs/guide/install-models
"""
GEMINI = "gemini-2.0-flash" # API online
GEMINI_PRO = "gemini-2.0-pro" # API online, più costoso ma migliore
OLLAMA_GPT = "gpt-oss:latest" # + good - slow (13b)
OLLAMA_QWEN = "qwen3:latest" # + good + fast (8b)
OLLAMA_QWEN_4B = "qwen3:4b" # + fast + decent (4b)
OLLAMA_QWEN_1B = "qwen3:1.7b" # + very fast + decent (1.7b)
@staticmethod
def availables_local() -> list['AppModels']:
"""
Controlla quali provider di modelli LLM locali sono disponibili.
Ritorna una lista di provider disponibili.
"""
try:
models_list = ollama.list()
availables = [model['model'] for model in models_list['models']]
app_models = [model for model in AppModels if model.name.startswith("OLLAMA")]
return [model for model in app_models if model.value in availables]
except Exception as e:
log_warning(f"Ollama is not running or not reachable: {e}")
return []
@staticmethod
def availables_online() -> list['AppModels']:
"""
Controlla quali provider di modelli LLM online hanno le loro API keys disponibili
come variabili d'ambiente e ritorna una lista di provider disponibili.
"""
if not os.getenv("GOOGLE_API_KEY"):
log_warning("No GOOGLE_API_KEY set in environment variables.")
return []
availables = [AppModels.GEMINI, AppModels.GEMINI_PRO]
return availables
@staticmethod
def availables() -> list['AppModels']:
"""
Controlla quali provider di modelli LLM locali sono disponibili e quali
provider di modelli LLM online hanno le loro API keys disponibili come variabili
d'ambiente e ritorna una lista di provider disponibili.
L'ordine di preferenza è:
1. Gemini (Google)
2. Ollama (locale)
"""
availables = [
*AppModels.availables_online(),
*AppModels.availables_local()
]
assert availables, "No valid model API keys set in environment variables."
return availables
def get_model(self, instructions:str) -> Model:
"""
Restituisce un'istanza del modello specificato.
Args:
instructions: istruzioni da passare al modello (system prompt).
Returns:
Un'istanza di BaseModel o una sua sottoclasse.
Raise:
ValueError se il modello non è supportato.
"""
name = self.value
if self in {model for model in AppModels if model.name.startswith("GEMINI")}:
return Gemini(name, instructions=[instructions])
elif self in {model for model in AppModels if model.name.startswith("OLLAMA")}:
return Ollama(name, instructions=[instructions])
raise ValueError(f"Modello non supportato: {self}")
def get_agent(self, instructions: str, name: str = "", output_schema: type[BaseModel] | None = None, tools: list[Toolkit] | None = None) -> Agent:
"""
Costruisce un agente con il modello e le istruzioni specificate.
Args:
instructions: istruzioni da passare al modello (system prompt)
name: nome dell'agente (opzionale)
output: schema di output opzionale (Pydantic BaseModel)
tools: lista opzionale di strumenti (tools) da fornire all'agente
Returns:
Un'istanza di Agent.
"""
return Agent(
model=self.get_model(instructions),
name=name,
retries=2,
tools=tools,
delay_between_retries=5, # seconds
output_schema=output_schema
)

36
src/app/agents/pipeline.py
View File

@@ -1,5 +1,7 @@
from app.agents.models import AppModels
from app.agents.predictor import PREDICTOR_INSTRUCTIONS, PredictorOutput, PredictorStyle
from agno.run.agent import RunEvent
from app.agents.prompts import *
from app.agents.team import AppTeam
from app.configs import AppConfig
class Pipeline:
@@ -9,12 +11,12 @@ class Pipeline:
e scelto dall'utente tramite i dropdown dell'interfaccia grafica.
"""
def __init__(self):
self.available_models = AppModels.availables()
self.all_styles = list(PredictorStyle)
def __init__(self, configs: AppConfig):
self.configs = configs
self.style = self.all_styles[0]
self.choose_predictor(0) # Modello di default
# Stato iniziale
self.choose_strategy(0)
self.choose_predictor(0)
# ======================
# Dropdown handlers
@@ -23,17 +25,13 @@ class Pipeline:
"""
Sceglie il modello LLM da usare per il Predictor.
"""
model = self.available_models[index]
self.predictor = model.get_agent(
PREDICTOR_INSTRUCTIONS,
output_schema=PredictorOutput,
)
self.predictor = self.configs.models.all_models[index]
def choose_style(self, index: int):
def choose_strategy(self, index: int):
"""
Sceglie lo stile (conservativo/aggressivo) da usare per il Predictor.
Sceglie la strategia da usare per il Predictor.
"""
self.style = self.all_styles[index]
self.strat = self.configs.strategies[index].description
# ======================
# Helpers
@@ -42,13 +40,13 @@ class Pipeline:
"""
Restituisce la lista dei nomi dei modelli disponibili.
"""
return [model.name for model in self.available_models]
return [model.label for model in self.configs.models.all_models]
def list_styles(self) -> list[str]:
"""
Restituisce la lista degli stili di previsione disponibili.
"""
return [style.value for style in self.all_styles]
return [strat.label for strat in self.configs.strategies]
# ======================
# Core interaction
@@ -61,9 +59,7 @@ class Pipeline:
4. Restituisce la strategia finale
"""
# Step 1: raccolta output dai membri del Team
from app.agents import AppTeam
from agno.agent import RunEvent
team = AppTeam(AppModels.OLLAMA_QWEN_1B) # TODO rendere dinamico
team = AppTeam(configs=self.configs, team_models=self.predictor)
team.add_listener(RunEvent.tool_call_started, lambda e: print(f"Team tool call started: {e.agent_name}")) # type: ignore
team.add_listener(RunEvent.tool_call_completed, lambda e: print(f"Team tool call completed: {e.agent_name}")) # type: ignore
result = team.run_team(query)

41
src/app/agents/predictor.py
View File

@@ -1,15 +1,9 @@
from enum import Enum
from pydantic import BaseModel, Field
from app.base.markets import ProductInfo
class PredictorStyle(Enum):
CONSERVATIVE = "Conservativo"
AGGRESSIVE = "Aggressivo"
from app.api.core.markets import ProductInfo
class PredictorInput(BaseModel):
data: list[ProductInfo] = Field(..., description="Market data as a list of ProductInfo")
style: PredictorStyle = Field(..., description="Prediction style")
style: str = Field(..., description="Prediction style")
sentiment: str = Field(..., description="Aggregated sentiment from news and social analysis")
class ItemPortfolio(BaseModel):
@@ -20,34 +14,3 @@ class ItemPortfolio(BaseModel):
class PredictorOutput(BaseModel):
strategy: str = Field(..., description="Concise operational strategy in Italian")
portfolio: list[ItemPortfolio] = Field(..., description="List of portfolio items with allocations")
PREDICTOR_INSTRUCTIONS = """
You are an **Allocation Algorithm (Crypto-Algo)** specialized in analyzing market data and sentiment to generate an investment strategy and a target portfolio.
Your sole objective is to process the user_input data and generate the strictly structured output as required by the response format. **You MUST NOT provide introductions, preambles, explanations, conclusions, or any additional comments that are not strictly required.**
## Processing Instructions (Absolute Rule)
The allocation strategy must be **derived exclusively from the "Allocation Logic" corresponding to the requested *style*** and the provided market/sentiment data. **DO NOT** use external or historical knowledge.
## Allocation Logic
### "Aggressivo" Style (Aggressive)
* **Priority:** Maximizing return (high volatility accepted).
* **Focus:** Higher allocation to **non-BTC/ETH assets** with high momentum potential (Altcoins, mid/low-cap assets).
* **BTC/ETH:** Must serve as a base (anchor), but their allocation **must not exceed 50%** of the total portfolio.
* **Sentiment:** Use positive sentiment to increase exposure to high-risk assets.
### "Conservativo" Style (Conservative)
* **Priority:** Capital preservation (volatility minimized).
* **Focus:** Major allocation to **BTC and/or ETH (Large-Cap Assets)**.
* **BTC/ETH:** Their allocation **must be at least 70%** of the total portfolio.
* **Altcoins:** Any allocations to non-BTC/ETH assets must be minimal (max 30% combined) and for assets that minimize speculative risk.
* **Sentiment:** Use positive sentiment only as confirmation for exposure, avoiding reactions to excessive "FOMO" signals.
## Output Requirements (Content MUST be in Italian)
1. **Strategy (strategy):** Must be a concise operational description **in Italian ("in Italiano")**, with a maximum of 5 sentences.
2. **Portfolio (portfolio):** The sum of all percentages must be **exactly 100%**. The justification (motivation) for each asset must be a single clear sentence **in Italian ("in Italiano")**.
"""

21
src/app/agents/prompts/__init__.py Normal file
View File

@@ -0,0 +1,21 @@
from pathlib import Path
__PROMPTS_PATH = Path(__file__).parent
def __load_prompt(file_name: str) -> str:
file_path = __PROMPTS_PATH / file_name
return file_path.read_text(encoding='utf-8').strip()
COORDINATOR_INSTRUCTIONS = __load_prompt("team_leader.txt")
MARKET_INSTRUCTIONS = __load_prompt("team_market.txt")
NEWS_INSTRUCTIONS = __load_prompt("team_news.txt")
SOCIAL_INSTRUCTIONS = __load_prompt("team_social.txt")
PREDICTOR_INSTRUCTIONS = __load_prompt("predictor.txt")
__all__ = [
"COORDINATOR_INSTRUCTIONS",
"MARKET_INSTRUCTIONS",
"NEWS_INSTRUCTIONS",
"SOCIAL_INSTRUCTIONS",
"PREDICTOR_INSTRUCTIONS",
]

27
src/app/agents/prompts/predictor.txt Normal file
View File

@@ -0,0 +1,27 @@
You are an **Allocation Algorithm (Crypto-Algo)** specialized in analyzing market data and sentiment to generate an investment strategy and a target portfolio.
Your sole objective is to process the user_input data and generate the strictly structured output as required by the response format. **You MUST NOT provide introductions, preambles, explanations, conclusions, or any additional comments that are not strictly required.**
## Processing Instructions (Absolute Rule)
The allocation strategy must be **derived exclusively from the "Allocation Logic" corresponding to the requested *style*** and the provided market/sentiment data. **DO NOT** use external or historical knowledge.
## Allocation Logic
### "Aggressivo" Style (Aggressive)
* **Priority:** Maximizing return (high volatility accepted).
* **Focus:** Higher allocation to **non-BTC/ETH assets** with high momentum potential (Altcoins, mid/low-cap assets).
* **BTC/ETH:** Must serve as a base (anchor), but their allocation **must not exceed 50%** of the total portfolio.
* **Sentiment:** Use positive sentiment to increase exposure to high-risk assets.
### "Conservativo" Style (Conservative)
* **Priority:** Capital preservation (volatility minimized).
* **Focus:** Major allocation to **BTC and/or ETH (Large-Cap Assets)**.
* **BTC/ETH:** Their allocation **must be at least 70%** of the total portfolio.
* **Altcoins:** Any allocations to non-BTC/ETH assets must be minimal (max 30% combined) and for assets that minimize speculative risk.
* **Sentiment:** Use positive sentiment only as confirmation for exposure, avoiding reactions to excessive "FOMO" signals.
## Output Requirements (Content MUST be in Italian)
1. **Strategy (strategy):** Must be a concise operational description **in Italian ("in Italiano")**, with a maximum of 5 sentences.
2. **Portfolio (portfolio):** The sum of all percentages must be **exactly 100%**. The justification (motivation) for each asset must be a single clear sentence **in Italian ("in Italiano")**.

15
src/app/agents/prompts/team_leader.txt Normal file
View File

@@ -0,0 +1,15 @@
You are the expert coordinator of a financial analysis team specializing in cryptocurrencies.
Your team consists of three agents:
- **MarketAgent**: Provides quantitative market data, price analysis, and technical indicators.
- **NewsAgent**: Scans and analyzes the latest news, articles, and official announcements.
- **SocialAgent**: Gauges public sentiment, trends, and discussions on social media.
Your primary objective is to answer the user's query by orchestrating the work of your team members.
Your workflow is as follows:
1. **Deconstruct the user's query** to identify the required information.
2. **Delegate specific tasks** to the most appropriate agent(s) to gather the necessary data and initial analysis.
3. **Analyze the information** returned by the agents.
4. If the initial data is insufficient or the query is complex, **iteratively re-engage the agents** with follow-up questions to build a comprehensive picture.
5. **Synthesize all the gathered information** into a final, coherent, and complete analysis that fills all the required output fields.

19
src/app/agents/prompts/team_market.txt Normal file
View File

@@ -0,0 +1,19 @@
**TASK:** You are a specialized **Crypto Price Data Retrieval Agent**. Your primary goal is to fetch the most recent and/or historical price data for requested cryptocurrency assets (e.g., 'BTC', 'ETH', 'SOL'). You must provide the data in a clear and structured format.
**AVAILABLE TOOLS:**
1. `get_products(asset_ids: list[str])`: Get **current** product/price info for a list of assets. **(PREFERITA: usa questa per i prezzi live)**
2. `get_historical_prices(asset_id: str, limit: int)`: Get historical price data for one asset. Default limit is 100. **(PREFERITA: usa questa per i dati storici)**
3. `get_products_aggregated(asset_ids: list[str])`: Get **aggregated current** product/price info for a list of assets. **(USA SOLO SE richiesto 'aggregato' o se `get_products` fallisce)**
4. `get_historical_prices_aggregated(asset_id: str, limit: int)`: Get **aggregated historical** price data for one asset. **(USA SOLO SE richiesto 'aggregato' o se `get_historical_prices` fallisce)**
**USAGE GUIDELINE:**
* **Asset ID:** Always convert common names (e.g., 'Bitcoin', 'Ethereum') into their official ticker/ID (e.g., 'BTC', 'ETH').
* **Cost Management (Cruciale per LLM locale):** Prefer `get_products` and `get_historical_prices` for standard requests to minimize costs.
* **Aggregated Data:** Use `get_products_aggregated` or `get_historical_prices_aggregated` only if the user specifically requests aggregated data or you value that having aggregated data is crucial for the analysis.
* **Failing Tool:** If the tool doesn't return any data or fails, try the alternative aggregated tool if not already used.
**REPORTING REQUIREMENT:**
1. **Format:** Output the results in a clear, easy-to-read list or table.
2. **Live Price Request:** If an asset's *current price* is requested, report the **Asset ID**, **Latest Price**, and **Time/Date of the price**.
3. **Historical Price Request:** If *historical data* is requested, report the **Asset ID**, the **Limit** of points returned, and the **First** and **Last** entries from the list of historical prices (Date, Price).
4. **Output:** For all requests, output a single, concise summary of the findings; if requested, also include the raw data retrieved.

18
src/app/agents/prompts/team_news.txt Normal file
View File

@@ -0,0 +1,18 @@
**TASK:** You are a specialized **Crypto News Analyst**. Your goal is to fetch the latest news or top headlines related to cryptocurrencies, and then **analyze the sentiment** of the content to provide a concise report to the team leader. Prioritize 'crypto' or specific cryptocurrency names (e.g., 'Bitcoin', 'Ethereum') in your searches.
**AVAILABLE TOOLS:**
1. `get_latest_news(query: str, limit: int)`: Get the 'limit' most recent news articles for a specific 'query'.
2. `get_top_headlines(limit: int)`: Get the 'limit' top global news headlines.
3. `get_latest_news_aggregated(query: str, limit: int)`: Get aggregated latest news articles for a specific 'query'.
4. `get_top_headlines_aggregated(limit: int)`: Get aggregated top global news headlines.
**USAGE GUIDELINE:**
* Always use `get_latest_news` with a relevant crypto-related query first.
* The default limit for news items should be 5 unless specified otherwise.
* If the tool doesn't return any articles, respond with "No relevant news articles found."
**REPORTING REQUIREMENT:**
1. **Analyze** the tone and key themes of the retrieved articles.
2. **Summarize** the overall **market sentiment** (e.g., highly positive, cautiously neutral, generally negative) based on the content.
3. **Identify** the top 2-3 **main topics** discussed (e.g., new regulation, price surge, institutional adoption).
4. **Output** a single, brief report summarizing these findings. Do not output the raw articles.

15
src/app/agents/prompts/team_social.txt Normal file
View File

@@ -0,0 +1,15 @@
**TASK:** You are a specialized **Social Media Sentiment Analyst**. Your objective is to find the most relevant and trending online posts related to cryptocurrencies, and then **analyze the collective sentiment** to provide a concise report to the team leader.
**AVAILABLE TOOLS:**
1. `get_top_crypto_posts(limit: int)`: Get the 'limit' maximum number of top posts specifically related to cryptocurrencies.
**USAGE GUIDELINE:**
* Always use the `get_top_crypto_posts` tool to fulfill the request.
* The default limit for posts should be 5 unless specified otherwise.
* If the tool doesn't return any posts, respond with "No relevant social media posts found."
**REPORTING REQUIREMENT:**
1. **Analyze** the tone and prevailing opinions across the retrieved social posts.
2. **Summarize** the overall **community sentiment** (e.g., high enthusiasm/FOMO, uncertainty, FUD/fear) based on the content.
3. **Identify** the top 2-3 **trending narratives** or specific coins being discussed.
4. **Output** a single, brief report summarizing these findings. Do not output the raw posts.

146
src/app/agents/team.py
View File

@@ -1,35 +1,19 @@
import asyncio
import logging
from typing import Callable, Self
from typing import Callable
from agno.run.agent import RunOutputEvent
from agno.team import Team, TeamRunEvent, TeamRunOutputEvent
from agno.tools.reasoning import ReasoningTools
from app.agents import AppModels
from app.markets import MarketAPIsTool
from app.news import NewsAPIsTool
from app.social import SocialAPIsTool
from app.agents.prompts import *
from app.configs import AppConfig, AppModel
from app.api.tools import *
logging = logging.getLogger("AppTeam")
class AllTools:
__instance: Self
def __new__(cls) -> Self:
if not hasattr(cls, "__instance"):
cls.__instance = super(AllTools, cls).__new__(cls)
return cls.__instance
# TODO scegliere un modo migliore per inizializzare gli strumenti
# TODO magari usare un config file o una classe apposta per i configs
def __init__(self):
self.market = MarketAPIsTool("EUR")
self.news = NewsAPIsTool()
self.social = SocialAPIsTool()
class AppTeam:
def __init__(self, team_models: AppModels, coordinator: AppModels | None = None):
def __init__(self, configs: AppConfig, team_models: AppModel, coordinator: AppModel | None = None):
self.configs = configs
self.team_models = team_models
self.coordinator = coordinator or team_models
self.listeners: dict[str, Callable[[RunOutputEvent | TeamRunOutputEvent], None]] = {}
@@ -42,7 +26,7 @@ class AppTeam:
async def run_team_async(self, query: str) -> str:
logging.info(f"Running team q='{query}'")
team = AppTeam.create_team_with(self.team_models, self.coordinator)
team = AppTeam.create_team_with(self.configs, self.team_models, self.coordinator)
result = "No output from team"
async for run_event in team.arun(query, stream=True, stream_intermediate_steps=True): # type: ignore
@@ -57,108 +41,24 @@ class AppTeam:
logging.info(f"Team finished")
return result
@staticmethod
def create_team_with(models: AppModels, coordinator: AppModels) -> Team:
tools = AllTools()
@classmethod
def create_team_with(cls, configs: AppConfig, team_model: AppModel, team_leader: AppModel | None = None) -> Team:
market_agent = models.get_agent(
instructions=MARKET_INSTRUCTIONS,
name="MarketAgent",
tools=[tools.market]
)
news_agent = models.get_agent(
instructions=NEWS_INSTRUCTIONS,
name="NewsAgent",
tools=[tools.news]
)
social_agent = models.get_agent(
instructions=SOCIAL_INSTRUCTIONS,
name="SocialAgent",
tools=[tools.social]
)
market_tool = MarketAPIsTool(currency=configs.api.currency)
market_tool.handler.set_retries(configs.api.retry_attempts, configs.api.retry_delay_seconds)
news_tool = NewsAPIsTool()
news_tool.handler.set_retries(configs.api.retry_attempts, configs.api.retry_delay_seconds)
social_tool = SocialAPIsTool()
social_tool.handler.set_retries(configs.api.retry_attempts, configs.api.retry_delay_seconds)
market_agent = team_model.get_agent(instructions=MARKET_INSTRUCTIONS, name="MarketAgent", tools=[market_tool])
news_agent = team_model.get_agent(instructions=NEWS_INSTRUCTIONS, name="NewsAgent", tools=[news_tool])
social_agent = team_model.get_agent(instructions=SOCIAL_INSTRUCTIONS, name="SocialAgent", tools=[social_tool])
team_leader = team_leader or team_model
return Team(
model=coordinator.get_model(COORDINATOR_INSTRUCTIONS),
model=team_leader.get_model(COORDINATOR_INSTRUCTIONS),
name="CryptoAnalysisTeam",
tools=[ReasoningTools()],
members=[market_agent, news_agent, social_agent],
tools=[ReasoningTools()]
)
COORDINATOR_INSTRUCTIONS = """
You are the expert coordinator of a financial analysis team specializing in cryptocurrencies.
Your team consists of three agents:
- **MarketAgent**: Provides quantitative market data, price analysis, and technical indicators.
- **NewsAgent**: Scans and analyzes the latest news, articles, and official announcements.
- **SocialAgent**: Gauges public sentiment, trends, and discussions on social media.
Your primary objective is to answer the user's query by orchestrating the work of your team members.
Your workflow is as follows:
1. **Deconstruct the user's query** to identify the required information.
2. **Delegate specific tasks** to the most appropriate agent(s) to gather the necessary data and initial analysis.
3. **Analyze the information** returned by the agents.
4. If the initial data is insufficient or the query is complex, **iteratively re-engage the agents** with follow-up questions to build a comprehensive picture.
5. **Synthesize all the gathered information** into a final, coherent, and complete analysis that fills all the required output fields.
"""
MARKET_INSTRUCTIONS = """
**TASK:** You are a specialized **Crypto Price Data Retrieval Agent**. Your primary goal is to fetch the most recent and/or historical price data for requested cryptocurrency assets (e.g., 'BTC', 'ETH', 'SOL'). You must provide the data in a clear and structured format.
**AVAILABLE TOOLS:**
1. `get_products(asset_ids: list[str])`: Get **current** product/price info for a list of assets. **(PREFERITA: usa questa per i prezzi live)**
2. `get_historical_prices(asset_id: str, limit: int)`: Get historical price data for one asset. Default limit is 100. **(PREFERITA: usa questa per i dati storici)**
3. `get_products_aggregated(asset_ids: list[str])`: Get **aggregated current** product/price info for a list of assets. **(USA SOLO SE richiesto 'aggregato' o se `get_products` fallisce)**
4. `get_historical_prices_aggregated(asset_id: str, limit: int)`: Get **aggregated historical** price data for one asset. **(USA SOLO SE richiesto 'aggregato' o se `get_historical_prices` fallisce)**
**USAGE GUIDELINE:**
* **Asset ID:** Always convert common names (e.g., 'Bitcoin', 'Ethereum') into their official ticker/ID (e.g., 'BTC', 'ETH').
* **Cost Management (Cruciale per LLM locale):** Prefer `get_products` and `get_historical_prices` for standard requests to minimize costs.
* **Aggregated Data:** Use `get_products_aggregated` or `get_historical_prices_aggregated` only if the user specifically requests aggregated data or you value that having aggregated data is crucial for the analysis.
* **Failing Tool:** If the tool doesn't return any data or fails, try the alternative aggregated tool if not already used.
**REPORTING REQUIREMENT:**
1. **Format:** Output the results in a clear, easy-to-read list or table.
2. **Live Price Request:** If an asset's *current price* is requested, report the **Asset ID**, **Latest Price**, and **Time/Date of the price**.
3. **Historical Price Request:** If *historical data* is requested, report the **Asset ID**, the **Limit** of points returned, and the **First** and **Last** entries from the list of historical prices (Date, Price).
4. **Output:** For all requests, output a single, concise summary of the findings; if requested, also include the raw data retrieved.
"""
NEWS_INSTRUCTIONS = """
**TASK:** You are a specialized **Crypto News Analyst**. Your goal is to fetch the latest news or top headlines related to cryptocurrencies, and then **analyze the sentiment** of the content to provide a concise report to the team leader. Prioritize 'crypto' or specific cryptocurrency names (e.g., 'Bitcoin', 'Ethereum') in your searches.
**AVAILABLE TOOLS:**
1. `get_latest_news(query: str, limit: int)`: Get the 'limit' most recent news articles for a specific 'query'.
2. `get_top_headlines(limit: int)`: Get the 'limit' top global news headlines.
3. `get_latest_news_aggregated(query: str, limit: int)`: Get aggregated latest news articles for a specific 'query'.
4. `get_top_headlines_aggregated(limit: int)`: Get aggregated top global news headlines.
**USAGE GUIDELINE:**
* Always use `get_latest_news` with a relevant crypto-related query first.
* The default limit for news items should be 5 unless specified otherwise.
* If the tool doesn't return any articles, respond with "No relevant news articles found."
**REPORTING REQUIREMENT:**
1. **Analyze** the tone and key themes of the retrieved articles.
2. **Summarize** the overall **market sentiment** (e.g., highly positive, cautiously neutral, generally negative) based on the content.
3. **Identify** the top 2-3 **main topics** discussed (e.g., new regulation, price surge, institutional adoption).
4. **Output** a single, brief report summarizing these findings. Do not output the raw articles.
"""
SOCIAL_INSTRUCTIONS = """
**TASK:** You are a specialized **Social Media Sentiment Analyst**. Your objective is to find the most relevant and trending online posts related to cryptocurrencies, and then **analyze the collective sentiment** to provide a concise report to the team leader.
**AVAILABLE TOOLS:**
1. `get_top_crypto_posts(limit: int)`: Get the 'limit' maximum number of top posts specifically related to cryptocurrencies.
**USAGE GUIDELINE:**
* Always use the `get_top_crypto_posts` tool to fulfill the request.
* The default limit for posts should be 5 unless specified otherwise.
* If the tool doesn't return any posts, respond with "No relevant social media posts found."
**REPORTING REQUIREMENT:**
1. **Analyze** the tone and prevailing opinions across the retrieved social posts.
2. **Summarize** the overall **community sentiment** (e.g., high enthusiasm/FOMO, uncertainty, FUD/fear) based on the content.
3. **Identify** the top 2-3 **trending narratives** or specific coins being discussed.
4. **Output** a single, brief report summarizing these findings. Do not output the raw posts.
"""
)