Farcaster agent (#142)

* Added first version of Farcaster agent

* Added placeholder for farcaster_client.py

* Agent posting on Farcaster

* formatting fixes

* Last fixes before review

* Last fixes before review (2)

* Added one PR comment

* Generating tweets from latest bets on Omen from think-thoroughly agent

* Adjusted prompts for 1st person

* Added twitter integration

* Fixing mypy

* Added lock file

* Implemented PR comments

* Fixed isort

* Fixed autoflake

* Added new PMAT version

* Formatting

* Updated lock file

* Updated lock file (2)

* Another round of PR comments

* Updated .env.example with Twitter credentials

.env.example

@@ -4,4 +4,10 @@ TAVILY_API_KEY=
 BET_FROM_PRIVATE_KEY=
 LANGFUSE_SECRET_KEY=
 LANGFUSE_PUBLIC_KEY=
-LANGFUSE_HOST=
+LANGFUSE_HOST=
+FARCASTER_PRIVATE_KEY=
+TWITTER_ACCESS_TOKEN=
+TWITTER_ACCESS_TOKEN_SECRET=
+TWITTER_BEARER_TOKEN=
+TWITTER_API_KEY=
+TWITTER_API_KEY_SECRET=

prediction_market_agent/agents/autogen_general_agent/README.md

@@ -0,0 +1,6 @@
+Farcaster agent URL
+https://warpcast.com/gnosis-ai-agent/
+
+FID (Farcaster ID) - 518226
+
+Public key - 0xF0AE06f9ca8bc3DC4C5C77CEC6F14D74AB21F782

prediction_market_agent/agents/autogen_general_agent/deploy.py

@@ -0,0 +1,62 @@
+from datetime import timedelta
+
+from loguru import logger
+from prediction_market_agent_tooling.config import PrivateCredentials
+from prediction_market_agent_tooling.deploy.agent import DeployableTraderAgent
+from prediction_market_agent_tooling.markets.data_models import Bet
+from prediction_market_agent_tooling.markets.markets import MarketType
+from prediction_market_agent_tooling.tools.utils import utcnow
+
+from prediction_market_agent.agents.autogen_general_agent.social_agent import (
+    build_social_media_text,
+)
+from prediction_market_agent.agents.autogen_general_agent.social_media.abstract_handler import (
+    AbstractSocialMediaHandler,
+)
+from prediction_market_agent.agents.autogen_general_agent.social_media.farcaster_handler import (
+    FarcasterHandler,
+)
+from prediction_market_agent.agents.autogen_general_agent.social_media.twitter_handler import (
+    TwitterHandler,
+)
+from prediction_market_agent.utils import APIKeys
+
+
+class DeployableSocialMediaAgent(DeployableTraderAgent):
+    model: str = "gpt-4-turbo-2024-04-09"
+    social_media_handlers: list[AbstractSocialMediaHandler] = [
+        FarcasterHandler(),
+        TwitterHandler(),
+    ]
+
+    def run(self, market_type: MarketType) -> None:
+        # It should post a message (cast) on each run.
+        # We just need one market to get latest bets.
+
+        bets = self.get_bets(market_type=market_type)
+        # If no bets available for the last 24h, we skip posting.
+        if not bets:
+            logger.info("No bets available from last day. No post will be created.")
+            return
+        tweet = build_social_media_text(market_type, bets)
+        self.post(tweet)
+
+    def get_bets(self, market_type: MarketType) -> list[Bet]:
+        better_address = PrivateCredentials.from_api_keys(APIKeys()).public_key
+        one_day_ago = utcnow() - timedelta(days=1)
+        return market_type.market_class.get_bets_made_since(
+            better_address=better_address, start_time=one_day_ago
+        )
+
+    def post(self, tweet: str | None) -> None:
+        if not tweet:
+            logger.info("No tweet was produced. Exiting.")
+            return
+
+        for handler in self.social_media_handlers:
+            handler.post(tweet)
+
+
+if __name__ == "__main__":
+    agent = DeployableSocialMediaAgent()
+    agent.deploy_local(market_type=MarketType.OMEN, sleep_time=540, timeout=180)

prediction_market_agent/agents/autogen_general_agent/prompts.py

@@ -0,0 +1,20 @@
+INFLUENCER_PROMPT = """
+        You are an AI agent that places bets on future events. 
+        You can find below a list of recent BETS that you have placed.
+
+        [BETS]
+        $BETS
+
+        Write one engaging tweet that attracts attention to the recent bets that the AI agent has placed, 
+        in order to increase his audience that wants to follow his betting activity. Make sure to include
+        the outcome that the AI agent has placed a bet on.
+        Pick a single topic for the tweet.
+        You must not add any reasoning or additional explanation, simply output the tweet.
+        """
+
+CRITIC_PROMPT = """
+        Reflect and provide critique on the following tweet. \n\n $TWEET
+        Note that it should not include inappropriate language.
+        Note also that the tweet should not sound robotic, instead as human-like as possible.
+        Also make sure to ask the recipient for an improved version of the tweet and nothing else.
+        """

prediction_market_agent/agents/autogen_general_agent/social_agent.py

@@ -0,0 +1,154 @@
+from datetime import datetime
+from enum import Enum
+from string import Template
+from typing import Any, Dict, Optional
+
+import autogen
+from autogen import AssistantAgent, UserProxyAgent
+from autogen.cache import Cache
+from prediction_market_agent_tooling.markets.data_models import Bet
+from pydantic import BaseModel
+
+from prediction_market_agent.agents.autogen_general_agent.prompts import (
+    CRITIC_PROMPT,
+    INFLUENCER_PROMPT,
+)
+from prediction_market_agent.utils import APIKeys
+
+
+class BetInputPrompt(BaseModel):
+    title: str
+    boolean_outcome: bool
+    collateral_amount: float
+    creation_datetime: datetime
+
+    @staticmethod
+    def from_bet(bet: Bet) -> "BetInputPrompt":
+        return BetInputPrompt(
+            title=bet.market_question,
+            boolean_outcome=bet.outcome,
+            collateral_amount=bet.amount.amount,
+            creation_datetime=bet.created_time,
+        )
+
+
+class AutogenAgentType(str, Enum):
+    WRITER = "writer"
+    CRITIC = "critic"
+    USER = "user"
+
+
+def reflection_message(
+    recipient: UserProxyAgent,
+    messages: list[Dict[Any, Any]],
+    sender: AssistantAgent,
+    config: Optional[Any] = None,
+) -> str:
+    reflect_prompt = Template(CRITIC_PROMPT).substitute(
+        TWEET=recipient.chat_messages_for_summary(sender)[-1]["content"]
+    )
+    return reflect_prompt
+
+
+def build_llm_config(model: str) -> Dict[str, Any]:
+    keys = APIKeys()
+    return {
+        "config_list": [
+            {"model": model, "api_key": keys.openai_api_key.get_secret_value()}
+        ],
+    }
+
+
+def build_agents(model: str) -> Dict[AutogenAgentType, autogen.ConversableAgent]:
+    llm_config = build_llm_config(model)
+
+    user_proxy = autogen.UserProxyAgent(
+        name="User",
+        human_input_mode="NEVER",
+        is_termination_msg=lambda x: x.get("content", "").find("TERMINATE") >= 0,
+        code_execution_config={
+            "last_n_messages": 1,
+            "work_dir": "tasks",
+            "use_docker": False,
+        },
+    )
+
+    writer = autogen.AssistantAgent(
+        name="Writer",
+        llm_config=llm_config,
+        system_message="""
+        You are a professional influencer, known for your insightful and engaging tweets.
+        You transform complex concepts into compelling narratives.
+        You should improve the quality of the content based on the feedback from the user.
+        You must always return only the tweet.
+        """,
+    )
+
+    critic = autogen.AssistantAgent(
+        name="Critic",
+        llm_config=llm_config,
+        system_message="""
+            You are a critic, known for your thoroughness and commitment to standards.
+            Your task is to scrutinize content for any harmful elements or regulatory violations, ensuring
+            all materials align with required guidelines.
+            References to betting and gambling are allowed.
+            """,
+    )
+
+    return {
+        AutogenAgentType.CRITIC: critic,
+        AutogenAgentType.WRITER: writer,
+        AutogenAgentType.USER: user_proxy,
+    }
+
+
+def build_social_media_text(model: str, bets: list[Bet]) -> str:
+    """
+    Builds a tweet based on past betting activity from a given participant.
+
+    This function utilizes the writer and critic agents to generate the tweet content. It first initializes the
+    necessary agents based on the provided model. Then, it registers a chat between the writer and critic agents.
+
+    The tweet content is generated using a template that includes questions about each market's title and likelihood.
+    """
+
+    agents = build_agents(model)
+    user_proxy, writer, critic = (
+        agents[AutogenAgentType.USER],
+        agents[AutogenAgentType.WRITER],
+        agents[AutogenAgentType.CRITIC],
+    )
+    user_proxy.register_nested_chats(
+        [
+            {
+                "recipient": critic,
+                "message": reflection_message,
+                "summary_method": "last_msg",
+                "max_turns": 1,
+            }
+        ],
+        trigger=writer,
+    )
+
+    # ToDO - fetch agent reasoning from DB and construct better tweets
+    #  See https://github.com/gnosis/prediction-market-agent/issues/150
+
+    task = Template(INFLUENCER_PROMPT).substitute(
+        BETS=[BetInputPrompt.from_bet(bet) for bet in bets]
+    )
+
+    # in case we trigger repeated runs, Cache makes it faster.
+    with Cache.disk(cache_seed=42) as cache:
+        # max_turns = the maximum number of turns for the chat between the two agents. One turn means one conversation round trip.
+        res = user_proxy.initiate_chat(
+            recipient=writer,
+            message=task,
+            max_turns=2,
+            summary_method="last_msg",
+            cache=cache,
+        )
+
+    tweet = res.summary
+    return str(
+        tweet
+    )  # Casting needed as summary is of type any and no Pydantic support

prediction_market_agent/agents/autogen_general_agent/social_media/abstract_handler.py

@@ -0,0 +1,10 @@
+import typing as t
+from abc import ABCMeta, abstractmethod
+
+
+class AbstractSocialMediaHandler(metaclass=ABCMeta):
+    client: t.Any
+
+    @abstractmethod
+    def post(self, text: str) -> None:
+        pass

prediction_market_agent/agents/autogen_general_agent/social_media/farcaster_handler.py

@@ -0,0 +1,19 @@
+from farcaster import Warpcast
+from loguru import logger
+
+from prediction_market_agent.agents.autogen_general_agent.social_media.abstract_handler import (
+    AbstractSocialMediaHandler,
+)
+from prediction_market_agent.utils import SocialMediaAPIKeys
+
+
+class FarcasterHandler(AbstractSocialMediaHandler):
+    def __init__(self) -> None:
+        api_keys = SocialMediaAPIKeys()
+        self.client = Warpcast(
+            private_key=api_keys.farcaster_private_key.get_secret_value()
+        )
+
+    def post(self, text: str) -> None:
+        cast = self.client.post_cast(text=text)
+        logger.info(f"Posted cast {cast.cast.text} - hash {cast.cast.hash}")

prediction_market_agent/agents/autogen_general_agent/social_media/twitter_handler.py

@@ -0,0 +1,27 @@
+import tweepy
+from loguru import logger
+from tweepy import Client
+
+from prediction_market_agent.agents.autogen_general_agent.social_media.abstract_handler import (
+    AbstractSocialMediaHandler,
+)
+from prediction_market_agent.utils import SocialMediaAPIKeys
+
+
+class TwitterHandler(AbstractSocialMediaHandler):
+    client: Client
+
+    def __init__(self) -> None:
+        keys = SocialMediaAPIKeys()
+
+        self.client = tweepy.Client(
+            keys.twitter_bearer_token.get_secret_value(),
+            keys.twitter_api_key.get_secret_value(),
+            keys.twitter_api_key_secret.get_secret_value(),
+            keys.twitter_access_token.get_secret_value(),
+            keys.twitter_access_token_secret.get_secret_value(),
+        )
+
+    def post(self, text: str) -> None:
+        response = self.client.create_tweet(text=text)
+        logger.info(f"Posted tweet {text} - response {response}")

prediction_market_agent/utils.py

@@ -42,6 +42,57 @@ def tavily_api_key(self) -> SecretStr:
         )
 
 
+class SocialMediaAPIKeys(APIKeys):
+    FARCASTER_PRIVATE_KEY: t.Optional[SecretStr] = None
+    TWITTER_ACCESS_TOKEN: t.Optional[SecretStr] = None
+    TWITTER_ACCESS_TOKEN_SECRET: t.Optional[SecretStr] = None
+    TWITTER_BEARER_TOKEN: t.Optional[SecretStr] = None
+    TWITTER_API_KEY: t.Optional[SecretStr] = None
+    TWITTER_API_KEY_SECRET: t.Optional[SecretStr] = None
+
+    @property
+    def farcaster_private_key(self) -> SecretStr:
+        return check_not_none(
+            self.FARCASTER_PRIVATE_KEY,
+            "FARCASTER_PRIVATE_KEY missing in the environment.",
+        )
+
+    @property
+    def twitter_access_token(self) -> SecretStr:
+        return check_not_none(
+            self.TWITTER_ACCESS_TOKEN,
+            "TWITTER_ACCESS_TOKEN missing in the environment.",
+        )
+
+    @property
+    def twitter_access_token_secret(self) -> SecretStr:
+        return check_not_none(
+            self.TWITTER_ACCESS_TOKEN_SECRET,
+            "TWITTER_ACCESS_TOKEN_SECRET missing in the environment.",
+        )
+
+    @property
+    def twitter_bearer_token(self) -> SecretStr:
+        return check_not_none(
+            self.TWITTER_BEARER_TOKEN,
+            "TWITTER_BEARER_TOKEN missing in the environment.",
+        )
+
+    @property
+    def twitter_api_key(self) -> SecretStr:
+        return check_not_none(
+            self.TWITTER_API_KEY,
+            "TWITTER_API_KEY missing in the environment.",
+        )
+
+    @property
+    def twitter_api_key_secret(self) -> SecretStr:
+        return check_not_none(
+            self.TWITTER_API_KEY_SECRET,
+            "TWITTER_API_KEY_SECRET missing in the environment.",
+        )
+
+
 def get_market_prompt(question: str) -> str:
     prompt = (
         f"Research and report on the following question:\n\n"

pyproject.toml

@@ -35,7 +35,7 @@ poetry = "^1.7.1"
 poetry-plugin-export = "^1.6.0"
 functions-framework = "^3.5.0"
 cron-validator = "^1.0.8"
-prediction-market-agent-tooling = { version = "^0.25.0", extras = ["langchain", "google"] }
+prediction-market-agent-tooling = { version = "^0.26.0", extras = ["langchain", "google"] }
 pydantic-settings = "^2.1.0"
 autoflake = "^2.2.1"
 isort = "^5.13.2"
@@ -53,6 +53,8 @@ pypdf2 = "^3.0.1" # TODO remove with https://github.com/gnosis/prediction-market
 faiss-cpu = "^1.8.0" # TODO remove with https://github.com/gnosis/prediction-market-agent/issues/97
 psycopg2-binary = "^2.9.9"
 sqlmodel = "^0.0.18"
+
+farcaster = "^0.7.11"
 mech-client = "^0.2.13"
 streamlit-extras = "^0.4.2"