RiskState

Introducing RiskState — risk governance for autonomous crypto agents. Your agent decides WHAT to trade. RiskState tells it HOW MUCH it can risk. → 30+ real-time data sources → 5-level policy engine (BLOCK → EXPANSION) → Deterministic & auditable (SHA-256) → BTC + ETH + DeFi lending aware → Any LLM, any framework, any execution layer Free beta API live → riskstate.ai

"But what if the dynamic system is wrong?" Every decision is deterministic. Same inputs → same policy → same SHA-256 hash. Full data provenance. No black box. You can always explain WHY the limit was what it was. Compare it to your current risk logic → riskstate.ai

Dynamic approach (RiskState): Panic + negative funding + high OI: → BLOCK: max 15%, reduce-only Confirmed trend + positive ETF flows + low leverage: → GREEN: max 60%, directional allowed Same agent. Same strategy. Radically different exposure. The environment dictates the limit, not a config file.

Static risk limits vs dynamic risk governance. Your agent's survival depends on which one it uses. 🧵

@NousResearch Congrats on v0.6.0. Profiles + MCP server mode is a big step for multi-strategy setups. We just published a reference architecture using Hermes as the orchestration layer for an autonomous BTC/ETH trading agent x.com/riskstate_ai/s…

The Hermes Agent update you've been waiting for is here.

@gkisokay Graeme, check out our risk governor for your local openclaw/hermes setup, we hope it helps, would love to hear your feedback

So the Virtuals stack is now: - deploy your own agent - use the Virtuals ecosystem to make your agent smarter - earn rewards for building smart agents I'll be testing this out to compare it to my local openclaw/hermes setup, but seems like a no brainer for new builders.

@sudoingX Check this setup for crypto agents: Hermes + CoinGlass (intelligence) + RiskState (risk governor) + CoW (execution) 👌🏻

what agent harness are you using and why? drop your reasoning below. lets find out what's keeping you on your current setup or what made you switch.

@instaclaws @virtuals_io You will need a policy engine to be able to win, check our risk governor out

want to make money with AI trading but don't have a Mac Mini or any technical experience? this is your sign. @virtuals_io is putting $100K/week on the line betting AI can't make money trading. zero risk to you, losses are backed by Virtuals. all you need is an instaclaw agent. deploy one in 60 seconds at instaclaw.io - no hardware, no code, no setup. your agent comes with a wallet, trading skills, and everything it needs to start competing. go prove them wrong! 🦞

@sharbel Without a risk governor, the bot overtrades in low-quality setups, plays oversized positions in weak conditions and uses leverage when the market doesn't support it. Check us out! and btw save a lot of money in LLM credits

how i built a trading bot with zero code: 345 wins. 26 losses. 93% win rate. no python knowledge. no stackoverflow. just me talking to an AI agent and iterating. here's the playbook: 1. start with the dumbest version possible. my first prompt was literally "build me a bot that bets on polymarket 15-minute markets" lost $11 on the first trade. but it worked. 2. iterate with real data. every time the bot lost money, i'd paste the trade log into claude and say "why did this lose? fix it." each loss = one prompt = one fix. 3. let the AI audit itself. set up a cron job that runs after every loss. reads the log, analyzes what went wrong, suggests fixes. the AI debugging itself. 4. brain + muscles trick. expensive model for complex decisions. cheap model for monitoring. free rule-based logic for actual trades. went from $300/month to $30/month in AI costs. 5. log everything. every trade, every decision, every loss with full forensics. when i say 345 wins that's from my actual log file. the bot runs 24/7 on a $600 mac mini. trades BTC, ETH, SOL every 15 minutes. the whole point: you don't need to know how to code. you need to know how to operate. describe what you want. test it with real money (small). paste the failures back in. iterate. the bot isn't smart. the process is.

@noisyb0y1 > you don’t have a policy engine, check us out

Automated Claude Bot system that will control my trades on Polymarket this past month i lost a lot of money trading predict market I asked AI agents why, and they all said the same thing > you don't have a clear strategy > you don't follow simple math formulas > you make emotional hype trades it built me a special assistant that helps with trading and stops me from making losing trades

@Bradgohtrades Definitely need a risk governor, check us out

I just lost $135k trading LIVE yesterday. I made the one fatal mistake of getting attached to the trade and not cutting the loss earlier. Watch the full breakdown here:

@fdotinc A deterministic policy engine that tells AI agents how much risk is allowed 👉🏻riskstate.ai/demo

We want to invest $100k-$200k in someone international. Show us a demo of what you're building, if it's cool: You'll join Canopy, a 5 week program: > can join in person at our SF campus or online > on demand access to our team + $50k in credits. > work w/ 500 founders making the same push. > extremely direct support on what you're shipping. > constant talks and lectures from $1b founders. not in SF? good. apply.

This is what your AI agent sees before every trade. One API call. Full risk enforcement. $ curl -X POST riskstate.netlify.app/v1/risk-state \ -H "Authorization: Bearer $KEY" \ -d '{"asset":"BTC"}' → policy_level: 3 (CAUTIOUS) → max_size_fraction: 0.35 → max_leverage: "1.5x" → blocked_actions: ["LEVERAGE_GT_2X"] → confidence_score: 0.72 💻See it live → riskstate.ai/demo

RiskState is now discoverable across the AI agent ecosystem. 🔌 MCP server npm — @riskstate/mcp-server awesome-mcp-servers Glama AI — AAA score mcp.so LobeHub 📄 SKILL.md SkillsMP SkillHub.club ClawHub.ai agentskills.io skills.sh — Vercel registry ➕REST API for everything else. Trading agents already know how to execute. Most still don't know when risk should be capped. One API call → policy level, max size, leverage caps, blocked actions. BTC + ETH. Free beta → riskstate.ai

clean architecture. the Risk Monitoring template is a good start but it's reactive — alerts after limits are breached. what about a policy node that sits before the Exchange Order and enforces constraints in real time? "max 25% size, no leverage above 2x, block trades when volatility is extreme" — based on live market conditions, not just portfolio metrics. would love to explore this as an integration.

@tom_doerr interesting framework — prompts as weights, Sharpe as loss function is a clean abstraction. the self-improvement loop handles strategy. what handles risk constraints? when the agent spawns new agents in different regimes, who caps their exposure?

Self-improving AI trading agents github.com/chrisworsey55/…

@TradeBoba solid stack. curious how risk limits work when automations run overnight — is there a policy layer capping exposure based on live market conditions, or is it fully user-defined?

Introducing Boba Agents, now live with free credits. We built OpenCode for trading - no APIs needed, access real market data for millions of tokens and RWAs. Trade smarter with semantic memory, automations, specialized agents, tool calls, and web search. Built for all models like Kimi 2.5, no setup required.

great piece. the yield curve modeling is smart — staying in the steep part of each curve instead of sliding down one is an insight most allocators miss. one question: the optimizer handles protocol-level risk (concentration, blast radius), but what governs market-level risk? if macro turns risk-off or on-chain metrics flash danger, does the optimizer reduce overall deployment or just reshuffle between protocols? that's the layer between "where to allocate" and "how much to allocate right now."

fascinating experiment. the simplification finding is real — most alpha gets buried under noise. one thing worth noting: fixed 8% position size across BTC, ETH and SOL treats all three as equally risky, which they're not. a dynamic risk layer that adjusts size based on live volatility, funding extremes, and macro regime would likely tighten that 0.3% MDD even further. also: the hold-out test will be the real moment of truth — 21.4 Sharpe on in-sample data is impressive but the overfitting question is still open.