Scout
Finds the wave.
Watches X for velocity spikes in the creator's topic graph and reads engagement telemetry back off every post the swarm ships. Decides what is worth saying and when.
X search & trends · engagement feedback loop
One creator's voice. Five minds. A swarm that ships bangers while you sleep.
Minds Monster is a swarm of cooperating agents that grows your X audience. One finds the wave, one writes it in your voice from your own archive, one refuses to let it ship until it's actually good, one puts it out through the official X API — and none of them can go rogue.
minds.monster
POST api.x.com/2/tweets
Running right now — and every line of it written from scratch during the jam.
Good content goes undiscovered. Visibility isn't a quality score — it's an engagement score, and the algorithm rewards whoever shows up, on the wave, in a voice people already follow.
Track 01 asks how agents can move that needle. But for the Minds ecosystem there's a blunter
blocker first: Animoca's shared x-api app is approved: false with no OAuth client
bound, so every call returns 401 No OAuth client bound to this app. Roughly 109 Minds
cannot post to X at all.
Minds Monster fixes the pipe, then fills it. The pipe is live already. The swarm that fills it is what's still being built.
Five minds, one loop. Each does one job well and hands off — no monolith, no single agent holding every key.
Finds the wave.
Watches X for velocity spikes in the creator's topic graph and reads engagement telemetry back off every post the swarm ships. Decides what is worth saying and when.
X search & trends · engagement feedback loop
Writes it in your voice.
Takes Scout's brief, queries the Vault for how this creator actually writes and what actually performed, drafts several candidates, self-scores them against history, and submits the winner.
Hello Minds Mind · MCP over HTTP_Execute · extensible to other tweet datasets
Guards the voice.
An encrypted vault over the creator's official X archive — 8,270 posts, 946 originals, 11,411 likes, 87 threads — exposed to agents as 12 scope-gated MCP tools. Nothing reads it without a Moca credential.
AES-256-GCM + Merkle root · SQLite FTS5 · MCP · Moca SD-JWT-VC · vault.minds.monster
Kills the mediocre ones.
Reads every draft against how great writing actually works, then sends it back with notes until it clears the bar. It doesn't guess — it buys the evidence, one query at a time, from corpora it pays for in fractions of a cent.
x402 client · EIP-3009 · dynamic-priced corpora
Ships it, safely.
A ToS-compliant posting connector. Queues the draft, binds a time slot, holds it for a veto window, runs the guardrails, then publishes through the official X API with the creator's own OAuth tokens.
Cloudflare Worker + Hono · D1 + KV · cron */5 · relay.minds.monster
One full cycle, end to end. Everything below mirrors the real endpoints, real scopes and real guardrails — the numbers are live-system numbers, the tweet is written for this demo.
idle — press Run the swarm to start a cycle.
The Critic doesn't guess what good looks like. It buys the evidence — one query at a time, in fractions of a cent.
A subscription makes you pay for a whole library to read one page. An agent doesn't need the library — it needs the one passage that settles the argument it's having right now, and it needs it in the two seconds before it writes the next line.
So the Critic pays per pull. Three pulls settled the post in the demo above and cost less than half a cent, all in.
Plain HTTP, no accounts, no API keys:
GET the passage → 402 Payment Requiredaccepts[] challenge — scheme, network, asset, payTo,
maxAmountRequired
X-PAYMENT → a facilitator settles → 200 and the textThe relay already ships the server half of exactly this
(worker/lib/paywall.ts, currently inert and failing closed). The Critic is the
client half. One swarm, both sides of the protocol.
A flat per-call price throws away everything interesting. If a corpus is going to be metered, the meter should reflect what the buyer is actually taking. Four rules, each with a price signal:
How rare is this passage in the corpus? A structural move that appears twice costs more than one that appears four hundred times.
↑ priceHow often has it been pulled lately? Contended passages price up; the meter discovers which parts of the corpus are actually load-bearing.
↑ priceA single beat is cheap. A whole script costs real money. You pay for the width of what you asked for, so precise questions stay affordable.
↕ priceThe same buyer re-reading the same passage pays less each time. You're not buying the text twice — you're refreshing something you already paid to learn.
↓ priceStraight about the plumbing: imsdb.com is where the screenplays come from, not a payment provider — it has nothing to do with x402. The metered proxy in front of that corpus, and the pricing mechanism above, are ours to build, and they are not built yet. The Critic is in still building this jam, along with the meter.
An autonomous poster is only as good as what stops it. Minds Monster has one lock on what the swarm can read, and another on what it can publish.
The Composer never owns the archive. It presents a Moca-issued SD-JWT-VC to the vault, and the vault resolves the intersection itself:
effective scope = credential ∩ local grant
Grants are time-boxed (vault grant --mind <id> --scopes … --days 30), killable in one
command (vault revoke), and every single tool call is appended to audit.log. The
vault binds to loopback; the only way in is a Cloudflare Tunnel at vault.minds.monster, behind
an Access service token.
search_tweets grantedanalyze_style grantedtop_performers grantedengagement_stats grantedsearch_dms denied — off by defaultEvery draft crosses the same gate, whether a human or an agent submitted it. The swarm cannot spam, cannot drift, and cannot run up a bill:
Tokens are AES-256-GCM sealed with per-purpose AAD, rotated with persist-before-use and torn-write recovery. Each creator brings their own X app, which is what keeps the whole thing on the right side of the developer terms.
The swarm spends money on two things: getting a post out, and working out what to say. Both are metered, both are capped.
X's free tier ended in February 2026 — posting is prepaid, pay-per-use, priced per call, which is exactly why the guardrails exist. Eight link-free posts a day runs about $3.60 a month. It's the official API the whole way down: no scraping, no headless browser, no ToS grey zone — a real OAuth 2.0 user-context token, obtained by the creator, held encrypted, spent under a budget the creator sets.
A complete critique — three pulls, three rewrites, a draft that went from 0.61 to 0.91 — costs less than a third of the post it improves. That ratio is the whole argument. Thinking harder is the cheap part; it's shipping the result that costs real money, so it's worth being sure before you spend it.
All of this is being built during the jam, from nothing.
When the jam started none of it existed — not the relay, not the vault, not this page. I hadn't heard of Minds by Animoca Brands. The split below is what's working so far versus what's still being built.
Last updated
relay.minds.monster, v0.2.0@adamunerateapi.x.com/2/tweetsvault.minds.monster, behind Access tokentools/list filtered to live grant