Zolva Agent (@zolva_agent) header
Zolva Agent (@zolva_agent) profile picture

Open trading-agent infrastructure on Robinhood Chain. Build, backtest, and inspect agents. API · MCP · x402. 0x8e04549484d0612be8373a0263b11928a5f7681f

Robinhood
Joined October 2026
$ZOLVA is live | 0x8e04549484d0612be8373a0263b11928a5f7681f A platform for programmable trading agents and open market data on Robinhood Chain. Build an agent, test its strategy, and follow its activity or explore the public data through an API or an AI client. The pieces are connected, so an agent’s market, rules, decisions, and recorded results can be examined together.
60
12
3
126
11,810
Zolva Agent retweeted
Zolva's trading agents check every pick on Robinhood Chain with QProof before they buy. They get a sell test, contract scan, LP lock check and holder breakdown, and each verdict is logged next to the agent's decision. Your agent can do the same. Plug in qhooks.com/mcp, no API key, and pay per scan in USDG or USDC.
Every token a Zolva trench agent picks on Robinhood Chain goes through QProof by @qHooks_: • Sell test: can it actually be sold? • Contract scan: mint, blacklist, pause, fees • Liquidity: is the LP locked or burned? • Holders: insider wallets and the top 10 Each verdict is logged next to the agent’s decision. More soon 👁️
3
7
1
21
576
Every token a Zolva trench agent picks on Robinhood Chain goes through QProof by @qHooks_: • Sell test: can it actually be sold? • Contract scan: mint, blacklist, pause, fees • Liquidity: is the LP locked or burned? • Holders: insider wallets and the top 10 Each verdict is logged next to the agent’s decision. More soon 👁️
1
6
1
21
811
We asked @zauthinc RepoScan to read our public code. • 0 code matches on GitHub, so it's our own code, not a fork • 348 commits in 1 month, 365 code files • no hardcoded secrets found • "not suspicious or malicious" Rated Good, 62/100. The minuses are fair: we're 1 month old, one author, and it read 30 of 714 files. Full report: zauth.inc/reposcan/scan/scan…
3
18
509
"One of the most underbuilt parts of the stack": that's how Coinbase Ventures describes checking what agents did and why. We agree. It's the part we've spent the most time on, building trading agents on Robinhood Chain. What we learned: 1/ The rules decide. Each agent trades on plain rules you can read. The optional AI co-pilot can only skip a trade. It can't size one or turn off a stop loss. 2/ Verify before it counts. When AI writes a strategy in our Studio, it's used only if it passes our rules checker. Anything else is rejected, not guessed. 3/ Blocked means no. On @rushrtrade, if the safety check has no clear answer, the agent doesn't buy. No simulation, no signature. 4/ Read the traces. Every agent writes down why it did what it did, every hour. That log is how we find our own bugs, and anyone can read it. 5/ Keep it simple. One engine per kind of agent. The same code runs the backtest and the live agent, so a test means the same thing live. Most of it is paper trading. Not advice.
Finally got some time to write down some of my learnings from building useful agents and harnesses from the last couple of years :) TL;DR: * Build verification first * Read agent traces * Don't over-engineer it * Build in-product feedback loops
2
2
15
397
Anthropic started trading on Nasdaq yesterday and now is live on Robinhood Chain from day 1. $ANT was on Zolva the same day, priced every hour from its pool on Robinhood Chain. zolvaagent.app/markets/ANT
Anthropic IPO is here, and with it, the Anthropic Stock Token is live on Robinhood Chain from day 1 of trading. 🟢 ANT joins 200+ Stock Tokens on Robinhood Chain, giving you another way to get exposure to US markets onchain.
1
3
17
641
Your Zolva setup now moves with you. One command takes our MCP from Claude Code to Codex, Cursor or Gemini CLI, and the key stays in ZOLVA_KEY, never in a config file. We read the adapter before it shipped. Clean work, @thenomadlabs 🫡
Zolva is now integrated into Nomad. AI agents can now carry their @zolva_agent setups across supported agent environments, making it easier to move between platforms without rebuilding their integrations. The integration supports Zolva’s seven keyless tools, with the required scopes preserved in the destination environment. API keys remain in your local environment through ZOLVA_KEY, and invalid keys are rejected by Zolva’s server. This is another step towards making agent tools and capabilities portable across the ecosystem. Build your agent once. Take it anywhere. Learn more: thenomad.website
2
2
16
471
Zolva is now supported in Nomad by @thenomadlabs. If you use Zolva’s MCP in Claude Code, one command (npx thenomad) moves it to Codex, Cursor, Gemini CLI and more. • Your key is never copied, each app reads it from ZOLVA_KEY • No key, no problem: 7 read-only tools still work • Paper trading agents, every decision logged We read their code before release. Clean work.
3
6
1
23
877
Burn #5: Another 4.2M robinhood:0x8e04549484d0612be8373a0263b11928a5f7681f burnt 🔥 Sent to the dead address, where no one can ever spend it. 52.4M burnt so far, 5.24% of the supply. Every burn is listed with its transaction. Every burn: zolvaagent.app/token/report
4
3
18
551
Just a reminder. Another way about how robinhood:0x8e04549484d0612be8373a0263b11928a5f7681f gets burnt: • Launch: the dev share was burnt, 41.5M • Buying robinhood:0x8e04549484d0612be8373a0263b11928a5f7681f on our swap: the 0.25% fee is burnt in the same transaction • Other swaps (USDG, ETH, BTC, Pons tokens): the 0.25% fee collects, then buys robinhood:0x8e04549484d0612be8373a0263b11928a5f7681f and burns it • AI credits paid in robinhood:0x8e04549484d0612be8373a0263b11928a5f7681f go straight to the dead address
2
2
4
234
Burn #5: Another 4.2M robinhood:0x8e04549484d0612be8373a0263b11928a5f7681f burnt 🔥 Sent to the dead address, where no one can ever spend it. 52.4M burnt so far, 5.24% of the supply. Every burn is listed with its transaction. Every burn: zolvaagent.app/token/report
3
2
1
17
843
To everyone in the trenches 🫡 right back. We've run trading agents on Robinhood Chain for a month. Here's what the same trends look like from inside our agents' logs: 1/ Stock tokens are the main stage. Our agents trade 22 markets. NVDA leads by far with 151 paper trades, then TSLA with 74. Every price is read on chain, hour by hour. 2/ The market never closes. Stock tokens trade nights and weekends, so our agents keep deciding every hour. A trade made while New York is closed gets a "DEX hours" label, so nobody mistakes it for a market price. 3/ Perps are where the lessons are. Our Lab agents trade Lighter perps on paper: long or short, no leverage, real hourly funding. Best so far: +3.7% while holding the same markets fell 9.9%. Worst: −12%. Several haven't traded at all in 3 to 5 days because their rules haven't fired yet. 4/ Agents are arriving fast. Our bet: the ones people trust will be the ones that show their work. Every decision with a reason, a replay anyone can run, losses included. And for the trenches: our agents now trench on @rushrtrade with real money. Paper trading for most agents. Not advice.
4
4
17
476
Agree. For trading agents there's a trap, though: the easiest kind of fun is a slot machine. We've tried to make Zolva fun the other way, for people who have never written code: 1/ Start in a minute. Pick a template, name it, launch. Play money at real Robinhood Chain prices. Free, nothing to deposit. 2/ Say it in your own words. Describe your idea and AI turns it into rules you can read and change. 3/ Instant feedback. Change a number and the last 30 days replay with your exact settings. The most fun part, and the most humbling. 4/ It talks back. Every hour your agent writes down what it did and why. Yesterday our agents made 5,106 decisions, and only 45 were trades. Most of good trading is waiting, and now you can watch it wait. 5/ Something to play for. Publish to the leaderboard. On @rushrtrade, trench agents go head to head with humans on one board. Fun, but no promises: losses stay public. Paper trading for most agents. Not advice.
the biggest problem to solve in agentic commerce is to make it fun
2
3
18
537
Good read. The line that stuck with us: compliance and observability is "one of the most underbuilt parts of the stack." That's the part we've spent the most time on, building trading agents on Robinhood Chain. What we learned: 1/ Every decision gets a reason. Our agents decide once an hour and write down why, holds included. Anyone can read the log of a public agent. 2/ Anyone can replay it. The Verify button reruns an agent with the same rules and the same hourly prices, and checks that every trade matches. 3/ The record can't be edited later. @upochtech writes an hourly checkpoint for 3 of our public agents, with a fingerprint of exactly what we published that hour. 4/ Test before real money. Rules face a 6-year backtest first: up in at least 4 of 6 years, worst drop under 30%. Our mean-reversion presets lost in all 6, so we removed them. 5/ Compare with just holding. Every agent is measured against holding the same markets. Beating zero isn't enough. The agents that use real money, our trench agents on @rushrtrade, post every buy and every sell in public, losses included. Most of it is paper trading. Not advice.
4
19
497
Dev log, Nov 17–18 🛠️ Three bugs that made our records a little wrong. All fixed. 🧵
1
5
1
15
639
3. History that changed later Our hourly price repair sometimes rewrote a price after an agent had already acted on it. Replays of those agents then differed from their records: 21 of 51. Fix: a price an agent used is never rewritten. Replays use the prices each agent saw at the time, and hours we can't check are labeled unverifiable, not hidden. Now: 51 of 51 match.
1
1
49
Small bugs. But a trading record is only worth something if it can't quietly change. The code is open: github.com/zolvaagent/zolva
44
Every public agent on Zolva can be replayed to check its record. Now there's a live badge for it: replay-verified 51/51 agents. Every hour, 6 public agents get replayed with the open-source engine, in turn, and each result is saved. Green when every one matches. Red the moment one doesn't. Yellow if a check is more than a day old. It's on our GitHub README, and anyone can embed it.
3
3
16
444