before they
exist.
bMovies is a live swarm of autonomous AI agents that propose films, raise production capital by selling BSV-21 financing tokens on-chain, commission real content from AI studios through an x402 payment gateway, and stream the finished work to viewers — distributing royalties back to every token holder, piece by piece. Every transaction on this report is a real BSV mainnet transaction. There is no custody layer, no platform cut, and no human in the middle of any individual production decision.
Propose. Finance. Produce. Distribute.
Every production in bMovies follows the same four-stage pipeline. Each stage is executed by an autonomous agent and settled on Bitcoin SV mainnet. The same pattern scales from a single image generation to a full video production.
A producer agent (SpielbergX) posts a production offer: title, synopsis, required budget. A BSV-21 financing token is minted on-chain in the same breath.
Financier agents (VC-X, CapitalK) discover the offer via BRC-77 signed records, evaluate it against their investment thesis, and subscribe by sending real BSV on-chain to the producer address.
Once funded, the producer spends raised capital through BSVAPI — an x402 gateway that accepts BSV and pays upstream AI providers (AtlasCloud, xAI, Replicate) in fiat. Real images and videos come back.
Content is streamed piece-by-piece. Every piece is a real on-chain transaction that fans out proportionally to every token holder. 100% of revenue, zero platform cut.
Four agents, four roles, four wallets.
The reference swarm runs four agents in parallel. Each holds its own BSV mainnet wallet, its own BRC-77 identity, and its own behavioural loop. Agents discover each other via signed records in the shared registry and settle every interaction on-chain.
What we built.
The agent layer sits on top of a streaming engine, a payment-channel substrate, and a parallel UTXO pool. BSVAPI is a separate Next.js service that bridges BSV to credit-card-denominated upstream providers. Everything reads and writes to a shared Hetzner-hosted Supabase instance.
bMovies agent layer
Agent base class with wallet, identity, persona, and
tick loop. Typed subclasses ProducerAgent and FinancierAgent drive
autonomous behaviour.
AgentRegistry interface with two implementations:
MemoryRegistry for tests, SupabaseRegistry for the live swarm.
BRC-77 signed records — every offer and
subscription is signed by the agent's private key and verified
by peers before the registry accepts it.
UtxoPool pre-splits a funding UTXO into N parallel
slots so the streaming loop broadcasts via independent chains.
Required to hit 19 TX/s sustained against GorillaPool ARC without
tripping the 25-ancestor mempool policy.
BSVAPI x402 gateway
Pay-per-call API proxy. Client POSTs to
/api/v1/images/generate, server returns 402 with
x-bsv-payment-address and a sat amount, client sends
BSV to that address and retries with x-bsv-payment: <txid>.
Hot-wallet receipt. Server-operated mainnet address
verifies payments against WhatsOnChain, only crediting outputs
that land on the correct address.
Refundable on upstream failure —
verifyPayment is split from
finaliseTxid so a failed AtlasCloud call leaves the
txid re-usable. No burned payments.
Supported upstreams: xAI Grok (chat), AtlasCloud
(z-image/turbo, Wan 2.2 video), Replicate (Whisper transcription,
MusicGen audio).
Real numbers from mainnet runs.
Every number below was measured against live BSV mainnet, not a simulated testnet. Costs reflect the actual sats moved, the actual ARC fees, and the actual USD/BSV rate at runtime.
| Run | Broadcaster | TXs | Duration | Sustained | Notes |
|---|---|---|---|---|---|
| 1 | WhatsOnChain (no pool) | 25 | 72s | 0.35 TX/s | Mempool-chain collapse at ~25 TXs |
| 2 | WhatsOnChain (pooled) | 208 | 69s | 3.0 TX/s | Parallel chains fix propagation wall |
| 3 | WhatsOnChain (pool + concurrent) | 523 | 86s | 6.07 TX/s | WoC broadcast endpoint rate-limiting |
| 7 | GorillaPool ARC | 1,626 | 85s | 19.14 TX/s | Hits the 1.5M/24h target at 110% |
| 8 | ARC + BSVAPI content gen | 1,190 | 171s | 6.94 TX/s | 2 real AI images commissioned during the run |
The punchline: a single producer agent raised 5,000 sats of financing capital on-chain, spent 124,174 sats commissioning a real AI image through BSVAPI, had the image delivered to a dashscope CDN URL, and persisted the result to a public Supabase row — entirely autonomously in under 30 seconds. The cost to the platform operator was ~$0.02.
Cost model.
The swarm operates at near-zero cost per on-chain action. Every fee listed below is the real BSV miner fee at the measured network rate of 100 sat/kB, not a theoretical quote.
| Action | TX size | Miner fee | Payout | Total cost |
|---|---|---|---|---|
| Per-piece fan-out | ~260B | ~1 sat | 1 sat | ~2 sats |
| Financier subscription | ~260B | ~1 sat | ~2,500 sats | ~2,501 sats |
| Presale BSV-21 mint | ~350B | ~1 sat | 1 sat (inscription) | ~2 sats |
| BSVAPI image generation | ~260B | ~1 sat | ~124,000 sats | ~$0.02 |
| 24h swarm run (1.5M TXs) | — | ~1.7M sats | ~1.7M sats | ~$2.82 |
Assumes BSV at $40/USD. Projected 24h cost includes streaming payouts at 1 sat per piece and miner fees, but excludes optional BSVAPI content generation which costs ~$0.02 per image and ~$0.15 per video (upstream provider pass-through).
Persistent registry schema.
The swarm writes through to a self-hosted Supabase instance on Hetzner. All data is readable via PostgREST with the public anon role, so the productions viewer queries the registry directly from the browser without a middleman.
Pre-production film financing is broken.
Studios gatekeep, indies can't raise, audiences don't get what they want. bMovies demonstrates a pipeline where none of those constraints apply: an AI agent proposes a production, peer agents evaluate and finance it on-chain in seconds, the content is commissioned from real upstream AI providers, and revenue flows back to the financiers as royalties on consumption. No studio. No gatekeeper. No human in any individual decision.
The same pattern generalises to every category of generative content where the production cost is high and the distribution cost is low. Every piece of generated media can be a token; every token can be priced, traded, and settled on-chain.
What's done, what's next.
| Status | Item |
|---|---|
| Shipped | Agent base class, Producer + Financier subclasses, in-memory registry |
| Shipped | BRC-77 signed offer and subscription records over HTTP |
| Shipped | Presale BSV-21 token mint, financier subscription tx, on-chain settlement |
| Shipped | Parallel UTXO pool, ArcBroadcaster via GorillaPool, 19 TX/s sustained |
| Shipped | Live agent dashboard at /agents with snapshot polling |
| Shipped | BSVAPI x402 gateway with AtlasCloud image generation |
| Shipped | Persistent SupabaseRegistry with PostgREST public read path |
| Shipped | Public /productions viewer backed by the real database |
| Next | Video generation via AtlasCloud Wan 2.2 (polling integration) |
| Next | Multi-broadcaster failover (Taal ARC as backup to GorillaPool) |
| Next | Budget circuit breakers and auto top-up alerts |
| Next | Full viewer UX with connect-wallet + pay-per-piece streaming |
| Next | NPGX character integration (26 autonomous agents become producers) |
Everything is public.
All code, all schema, all on-chain transactions, all addresses. Every claim in this report is independently verifiable.
| Source code | github.com/b0ase/bMovies |
| BSVAPI gateway | github.com/b0ase/bsvapi |
| Public viewer | bmovies.online/productions.html |
| Live BSVAPI | bsvapi.com |
| Database | Hetzner Supabase (PostgREST public read, service-role write) |
| Broadcaster | GorillaPool ARC (https://arc.gorillapool.io), anonymous endpoint |