AI Agents Suite
A modular collection of AI agents for automating repetitive tasks across social media, trading, moderation, and development workflows.
Overview
The Problem
Repetitive tasks drain time: scheduling social posts, moderating communities, monitoring markets, reviewing code. Each domain has its own tools, none of which talk to each other. Context switching kills productivity.
The Solution
Build once, deploy everywhere. A unified agent framework where each agent specializes in one domain but shares infrastructure: scheduling, logging, error handling, and a common LLM interface.
Architecture
┌─────────────────────────────────────────────────────────────────┐
│ AGENT ORCHESTRATOR │
│ Scheduling • Logging • Error Handling • Rate Limiting │
└─────────────────────────────────────────────────────────────────┘
│
┌──────────────────────┼──────────────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ SOCIAL AGENT │ │ SENTINEL AGENT │ │ TRADING AGENT │
│ Twitter/X │ │ Discord/Slack │ │ Market Data │
│ Content Gen │ │ Content Filter │ │ Signal Gen │
└─────────────────┘ └─────────────────┘ └─────────────────┘
│ │ │
└──────────────────────┼──────────────────────┘
▼
┌─────────────────────────────────────────────────────────────────┐
│ LLM INTERFACE │
│ OpenAI / LLM / Local Models (Ollama) │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ EXTERNAL APIs │
│ Twitter API • Discord API • Exchange APIs • GitHub API │
└─────────────────────────────────────────────────────────────────┘Orchestration
Central coordinator handles scheduling, retry logic, and cross-agent communication.
Specialized Agents
Each agent is a self-contained module with domain-specific prompts and tools.
LLM Abstraction
Swap between OpenAI, LLM, or local models without changing agent code.
Agent Types
Social Media Agent
Automated posting, engagement tracking, and AI-generated responses.
- • Multi-platform scheduling
- • Content generation with brand voice
- • Engagement analytics
Sentinel - AI Guardian
Autonomous agent that actively protects and informs your server.Click to view details →
- • The Cleaner: Auto-moderates and bans suspicious accounts instantly.
- • Hybrid Brain: Trainable via
/learncommands. Remembers context. - • News Anchor: Auto-posts crypto news.
- • Researcher: Browses web.
Trading Agent
Market monitoring and signal generation (analysis only, no execution).
- • Technical indicator monitoring
- • Sentiment analysis from social
- • Alert notifications
Dev Assistant
Code review, documentation generation, and CI/CD automation.
- • PR review suggestions
- • Auto-generated docs
- • Test generation
Tech Stack
Every tool was chosen for a specific reason:
Python
The lingua franca for AI/ML. Every LLM SDK has first-class Python support. Async/await handles concurrent API calls efficiently.
LangChain
Agent framework with built-in tools, memory, and chain composition. Accelerates development vs building from scratch.
FastAPI
High-performance Python API framework. Automatic OpenAPI docs, Pydantic validation, and async support out of the box.
Redis
Task queue for background jobs. Rate limiting per API. Caching LLM responses to reduce costs. Pub/sub for real-time events.
Challenges
LLM Cost Management
Problem: Agents making hundreds of API calls per day adds up quickly.
Solution: Implemented response caching in Redis. Semantic similarity check before new API call. Smaller models for simple tasks, larger for complex reasoning.
API Rate Limits
Problem: Twitter, Discord, and exchange APIs all have different rate limits.
Solution: Per-API rate limiters using Redis token buckets. Exponential backoff on 429 errors. Priority queuing for time-sensitive tasks.
Agent Reliability
Problem: LLMs sometimes produce unexpected outputs that break downstream logic.
Solution: Structured output with Pydantic validation. Fallback handlers for malformed responses. Human-in-the-loop for high-stakes actions.
Results
What Works Today
- • Core orchestrator with scheduling
- • Sentinel - AI Guardian (Discord)
- • Social agent prototype (Twitter)
- • LLM abstraction layer
- • Redis-based task queue
- • Basic logging and monitoring
Planned Next
- • Trading signal agent
- • Web dashboard for configuration
- • Multi-agent collaboration
- • Self-hosted local models