MENTAT

explain like i'm four

agents, made boring on purpose.

No mystique, no marketing. The eight things you actually need to understand before you train your first agent.

tldr

An agent is a chatbot that can do things, not just say things. Imagine a really good assistant who can also click buttons.

what makes it more than a chatbot

A normal LLM (like ChatGPT in the box) just talks. An agent has tools — small programs it can run. "Send 1 USDC to this address", "fetch this URL", "post this tweet". The brain decides; the tools act.

It's like the difference between a smart friend giving you advice on the phone, and a smart friend who is sitting at your desk and can actually open the laptop.

the loop

Every conversation goes you say something → agent thinks → agent picks a tool (maybe) → tool runs → agent reads the result → agent talks to you. Repeat. That's it. There is no magic.

why this matters

Once an agent can use tools, it can do real work — book the flight, send the message, file the expense, do the trade. A chatbot tells you what to do. An agent does it.

marketplace

courses, MCP packs, skills.

Everything Mentat sells, in one list. League One stays free forever. Specialist courses are one-time. MCP packs and skills are drop-in modules — bring them to your own agent or a Mentat Private slot.

courses

multi-day curricula. preset lessons, no LLM cost.

  • course

    League One

    free · always

    7-day onboarding. Zero to your first AI agent.

    Plain-English curriculum delivered web + Telegram. No code required. Daily quizzes and 24-hour pacing. Day 7 graduates ship their own IronClaw agent on the included starter credit.

    • click-through curriculum, no model access required to learn
    • voice + web + telegram, one account
    • day 7 = your first agent on IronClaw
    freestart free →

  • course

    Funding Arb Specialist

    live

    Delta-neutral basis trade. Hyperliquid + Nado.

    The strategy skyto runs every week, documented as a 5-day course. Funding-rate math, venue selection, sizing, paper-mode harness, kill-switches.

    • 5 days · zero-LLM lessons · 3 quiz qs + homework per day
    • paper-trade harness for week 1 before any real fills
    • lifetime access · all future revisions included
    $99$50 w/ grad
    see syllabus →

  • course

    On-Chain Reading & Indexing

    coming soon

    Helius, Alchemy, Etherscan, NEAR Indexer — into your agent.

    How to wire real-time on-chain feeds into an agent without melting the wallet. Webhook design, rate-limit pacing, indexer fallbacks, minimum-data RAG.

    • 4 days · indexer + webhook architecture
    • real Helius + Alchemy webhooks, real recovery paths
    • patterns reused across all on-chain deputies
    $79$40 w/ grad
    request preorder access

  • course

    Voice + Vision Specialist

    coming soon

    Talk to your agent. Show it a screenshot.

    Whisper STT, OpenAI TTS / xTTS local, screenshot OCR, photo classification routing into the brain. End-to-end voice loop on Telegram in three days.

    • whisper STT auto-on for telegram voice notes
    • TTS reply pipeline with 10 voice options
    • screenshot → OCR → tool call routing
    $69$35 w/ grad
    request preorder access

MCP packs

pre-built deputies you drop into any MCP-capable agent.

  • MCP pack

    Hyperliquid MCP Pack

    live

    Drop-in deputies for perps + spot.

    Pre-built MCP server exposing Hyperliquid deputy tools — get-funding-rate, calc-delta-neutral-size, open/close paper position, watch order book. Plug into any IronClaw or local agent.

    • 12 tools · paper + live mode
    • streaming order book, bbo, trades
    • works with any MCP-capable agent (Claude, IronClaw, local)
    $49$25 w/ grad
    request access →

  • MCP pack

    TON Cross-DEX Pack

    coming soon

    STON.fi ↔ DeDust arbitrage tooling.

    Tools for finding cross-DEX price spreads on TON, sizing safely against pool depth, executing the round-trip with a single confirmation, watching mempool for sandwich risk.

    • STON.fi + DeDust depth + price reads
    • mempool watcher catches sandwiches before they land
    • one-shot round-trip execution with confirm gate
    $79$40 w/ grad
    request preorder access

  • MCP pack

    Solana Whale Tracker Pack

    coming soon

    Helius webhooks → sub-second whale alerts.

    Watch any Solana wallet, fire Telegram alerts on threshold moves, auto-classify swap vs send vs LP, optionally copy-trade with size + slippage caps.

    • helius webhook handler with retry queue
    • classifier separates noise from real moves
    • copy-trade scaffold (paper-default)
    $59$30 w/ grad
    request preorder access

  • MCP pack

    NEAR DeFi Pack

    coming soon

    Ref Finance + Burrow in one bundle.

    Triangle arbitrage discovery on Ref, supply/borrow/repay on Burrow, NEAR validator search by reward + commission. Designed for skyto.near operators.

    • ref triangle arb scanner (paper-default)
    • burrow lend/borrow with health-factor gate
    • validator scoring by 30-day reward consistency
    $59$30 w/ grad
    request preorder access

skills + modules

single-purpose modules — confirmation gates, voice, memory.

  • skill

    Memory + Recall Module

    coming soon

    Two-tier durable memory + Qdrant recall.

    Drop into any agent: a markdown notebook for facts the agent should never forget, plus a Qdrant vector store for older conversation snippets the agent can search by meaning.

    • markdown notebook + qdrant vector store
    • auto-classifier decides what to persist
    • works with any python agent loop
    $19$10 w/ grad
    request access →

  • skill

    Confirmation Gate Wrapper

    coming soon

    Drop-in yes/no/phrase-gate for any tool.

    Decorator that turns any python function into a phrase-gated action. Persists pending confirmations to SQLite, surfaces a one-line summary on Telegram, supports paper-mode bypass.

    • one-line decorator: @phrase_gated("send")
    • pending confirmations table + replay-safe
    • paper-mode bypass; live requires explicit phrase
    $19$10 w/ grad
    request access →

  • skill

    Voice STT/TTS Wrapper

    coming soon

    Whisper + 10-voice TTS in a single import.

    Plug Whisper into your Telegram bot for incoming voice notes, plug OpenAI / xTTS for replies. Includes voice picker, autoplay logic, and a fallback chain when one provider degrades.

    • whisper STT + openai/xtts TTS
    • 10 voice options; per-user preference
    • graceful fallback if one provider 5xxs
    $29$15 w/ grad
    request access →

  • skill

    Self-Mod Sandbox

    coming soon

    Run agent-written code safely.

    A docker-network-isolated sandbox that runs whatever code your agent writes, executes the test suite inside it, and returns a structured pass/fail. Triple-rollback on apply (git + FS snapshot + DB diff).

    • docker isolation, no host network, no host secrets
    • test runner with structured pass/fail output
    • three-layer rollback if a deploy misbehaves
    $49$25 w/ grad
    request access →

  • skill

    Telegram Bot Boilerplate

    coming soon

    Tool-loop bot in 200 lines.

    A minimal Telegram bot template wired to a Claude / NEAR AI / Ollama loop with tool-use, confirmation gate, multi-user routing, voice in/out, and graceful Anthropic fallback. The starter kit Mentat itself grew out of.

    • claude tool-use loop ready to extend
    • multi-user routing with allowlist
    • graceful provider fallback chain
    $29$15 w/ grad
    request access →

prices in USD. Stripe checkout lands in a follow-up — for now paid items route through the operator request form.

also see

public feed

Graduates

What League One alumni have shipped on Day 8. Submit yours after you finish.

browse builds →

3000 items

Capability index

The full menu of things an agent could do. Wired vs in-scope. Pick a build target.

open /explore →

ready to start?

League One is free. 7 days. You don't need crypto, an API key, or your own server.

join league one →advanced docs