Agentarium Multi Framework Agents

agentarium-multi-framework-agents - Agentic AI

A hands-on collection of advanced Agentic AI systems built across multiple orchestration frameworks, including LangGraph, MCP, OpenAI Agents SDK, CrewAI, and AutoGen. Instead of focusing on one narrow use case, it demonstrates how autonomous agents can power real interactive products across finance, research, software engineering, gaming, storytelling, sports management, mystery-solving, and creator workflows, with persistent state, custom tool servers, rich Gradio interfaces, and local Ollama-based model, Gemini & Openrouter models execution.

🤖 Agents

Agent	Tech Stack	Summary	Agent Architecture & Features
AI Dungeon Master Agent	LangGraph, MCP (custom world-state server), Ollama, Gradio, JSON persistence	A persistent storyworld engine where a hidden LangGraph Dungeon Master resolves player actions through MCP tools for dice rolls, NPC registration, quest progression, inventory changes, and canonical scene updates before a narrator agent turns the updated state into cinematic playable prose with exactly three next-step choices; supports Fantasy, Sci-Fi, Cyberpunk, Post-Apocalypse, and Mythic Horror campaigns with saved session loading in a Gradio UI.	Architecture & Features
SWE Team Agent	CrewAI, Gemini 2.5 Flash / Ollama, pytest	A sequential 5-agent software engineering crew that takes plain-English requirements and produces a complete, runnable codebase — engineering lead writes a detailed design, backend engineer implements it, a dedicated code reviewer audits for bugs and edge cases, the backend engineer applies targeted fixes, a frontend engineer builds a Gradio UI, and a test engineer writes full unit tests — then automatically runs pytest and consolidates everything into a build summary report.	Architecture & Features
Detective Case Files Agent	LangGraph, MCP (custom detective case server), Ollama, Gradio, JSON persistence	A persistent interactive mystery board where a hidden LangGraph investigator inspects evidence, interviews suspects, compares alibis, ranks suspicion, and updates a saved theory board before presenting the user with a polished detective briefing; supports penthouse murder, museum heist, and cyber sabotage cases with reloadable sessions in a Gradio UI.	Architecture & Features
AI Equity Traders Agent	OpenAI Agents SDK, MCP (custom accounts server + Polygon market + Serper + mcp-server-fetch + mcp-memory-libsql + Telegram push), Ollama / multi-model, SQLite, Gradio, Plotly	A production-grade autonomous trading simulation where 5 AI traders (Warren, George, Ray, Cathie, Ed) run concurrently in a timed loop — each backed by a Researcher sub-agent with a persistent per-trader knowledge graph, live web search, and market data; traders alternate between trade and rebalance cycles, log every action via a custom OpenAI tracing hook, fire structured Telegram alerts per trade, and are guarded by a drawdown circuit breaker that halts any trader exceeding a configurable loss threshold; a Gradio UI streams live logs, portfolio charts, holdings, transactions, and a real-time leaderboard.	Architecture & Features
Fantasy Sports GM Agent	LangGraph, MCP (custom fantasy league server), Ollama, Gradio, JSON persistence	A persistent fantasy team manager where a hidden LangGraph GM reads a saved roster and matchup state, computes safe/balanced/upside lineups, scouts waiver wire targets, proposes trade packages, and saves a complete weekly plan before presenting the user with a polished briefing; supports football, basketball, and cricket with reloadable sessions in a Gradio UI.	Architecture & Features
Business Agent Creator Agent	AutoGen Core, gRPC distributed runtime, OpenRouter (5 free models), Ollama (qwen3:8b), Gradio	A self-expanding distributed multi-agent system where a Creator agent generates brand-new AI agents at runtime — writing their Python code to disk, dynamically importing and registering them into the live gRPC runtime, then sending them their first task; each generated agent has a unique sector and personality, randomly bounces ideas to peers for refinement, and runs on a different free OpenRouter model; an Evaluator agent then scores all ideas on originality, market size, feasibility, and AI centrality, producing a ranked leaderboard — with multi-round idea evolution, persistent archive with deduplication, and a Gradio UI.	Architecture & Features
Creator Studio Agent	LangGraph, MCP (custom creator studio server), Ollama, Gradio, JSON persistence	A persistent content war room where a hidden LangGraph showrunner scans a niche trend radar, builds a content concept, drafts titles and thumbnail direction, creates the publish plan, and saves a greenlit package before presenting the user with a polished creator brief; supports gaming, AI, sports, and lifestyle channels across YouTube, TikTok, X, and Instagram in a visual dashboard UI.	Architecture & Features
SDR Agent	OpenAI Agents SDK, Ollama, Resend API	Runs 3 persona-specialised email writers (professional, engaging, concise) in parallel against a personalised prospect brief, scores all 3 with a critic agent on clarity/urgency/personalization, sends the winner as a styled HTML email, then auto-generates a 3-message Day-3/7/14 follow-up cadence — with batch mode for prospect lists and a local JSON audit log.	Architecture & Features
Deep Research Agent	OpenAI Agents SDK, WebSearchTool (gpt-4o-mini), Ollama (planner/writer/critic), Resend API, Gradio	Plans a multi-angle web search strategy, runs all searches in parallel with source URL tracking, synthesises a 1000+ word cited markdown report, then self-critiques on Depth/Accuracy/Coherence and rewrites if any score falls below 6 — before emailing the final report, saving it to a local archive, and surfacing clickable follow-up research topics in the Gradio UI.	Architecture & Features
Stock Picker Agent	CrewAI, Ollama (qwen3:8b + deepseek-r1:8b), SerperDev, ChromaDB, SQLite, Telegram Bot API	Hierarchical multi-agent crew that finds 2-3 trending companies in any sector via live web search, deep-researches each one, then passes the top candidate to a dedicated risk analyst that scores it on volatility, regulatory exposure, and competitive moat — before the stock picker selects the best risk-adjusted opportunity and fires a rich formatted Telegram alert. Supports multi-sector parallel runs and logs every pick to a persistent history file.	Architecture & Features
Career Conversation Agent	OpenAI-compatible LLM (Ollama), Gradio, Telegram Bot API, PyPDF	Impersonates a professional persona on a personal website — reads a LinkedIn PDF and bio to answer career questions in character, auto-classifies visitors as recruiters or clients to adapt tone, captures leads with email and meeting requests, and fires structured real-time Telegram alerts for every lead, unanswered question, and booking.	Architecture & Features
Sidekick Operator Agent	LangGraph, LangChain, Playwright, Ollama, Gradio, Telegram Bot API	A persistent personal co-worker built on a LangGraph worker-evaluator loop — the worker uses Playwright browser, live web search, Python REPL, Wikipedia, weather, and file tools to complete any task, while a structured-output evaluator scores each attempt against user-defined success criteria and loops back with targeted feedback until the bar is met or user input is needed; enforces a configurable max-step guard, surfaces a full tool audit trail per task, and logs every completed task to a persistent history file accessible from a dedicated Gradio tab.	Architecture & Features
Financial Researcher Agent	CrewAI, Ollama (qwen3:8b + deepseek-r1:8b), SerperDev, Gradio	A sequential 3-agent financial research crew that uses live web search to gather current financials, news, and competitive positioning for any company, passes the findings to an analyst who writes a tailored investment-grade report personalised to the user's profile from a knowledge file, then runs a dedicated fact-checker that audits every claim for accuracy and flags unverified figures — with comparative multi-company analysis mode, think-tag stripping, research history log, and a Gradio UI.	Architecture & Features
Rock Paper Scissors Agent	AutoGen Core, SingleThreadedAgentRuntime, Ollama (qwen3:8b + llama3.2:3b), Gradio	Teaches AutoGen Core's message-routing runtime through a progressively complex Rock Paper Scissors implementation — strategy-aware player agents (aggressive, adaptive, defensive, random) with validated choice extraction and retry logic, a multi-round orchestrator with live scoreboard and Pydantic structured round results, a full match summary with narrative recap, human-vs-agent mode, and a Gradio UI with single-round, series, and human-vs-AI tabs.	Architecture & Features

Gradio UI for Some Agents

AI Dungeon Master Agent

---

Detective Case Files Agent

---

AI Equity Traders Agent

---

Fantasy Sports GM Agent

---

Creator Studio Agent

---

Other Agents

Business Agent Creator Agent	Financial Researcher Agent

---

Rock Paper Scissors Agent	SWE Team Agent

Credits

• Agentic AI By Ed Donner -- Initial versions of some agents are taken from here and upgraded on top of that
• Claude (Free Tier)
• ChatGPT GO
• Codex