indie.money
Marketplace for n8n-compatible AI agents, paid per run
What it is: indie.money — you publish an n8n-compatible agent, set a purchase price + per-run fee, and sell it. Three personas: Builders create and price agents, Producers license and resell them through a hosted chat UI, Buyers pay per run. Beta opened this month; GA in September. Base mainnet.
The economics (why I built it this way):
- Builders keep 100% of purchase, per-run, and subscription revenue. The only platform cut is 10% when you cash out Earnings → USDC. (There's a flat $0.001 payment-processing fee per pay-as-you-go run — that goes to the payments provider, not us. I won't call it "no fees," because it isn't.)
- AI/API cost is billed to the caller, not the builder — with a 0–300% builder markup. So model cost is a profit lever, not your COGS. A producer reselling a $0.05/run agent at $5 keeps $4.95/run; the AI is the caller's bill.
- Buyers need no crypto — email/Google sign-in. Payments are pay-as-you-go and settle on Base via x402 under the hood (caller-pays, prepaid, or producer-funded rails).
Why now: Apify just shipped x402 payments for 20,000+ Actors, and Coinbase launched Agentic.Market — so "agents pay per run" was just proven at 20K-tool scale. That part is no longer a bet. But Apify sells their catalog: you list into it, they take a cut, and prices run through their credits. indie.money is the builder-owned version — you publish your own agent, set your own price, and keep 100% of it — 10% only when you cash out — with a resale layer (Producers) on top and a per-run track record that compounds on something you own. That's the wedge.
Where it's rough: it's a beta. Small catalog, early UX, Base-only. I'm looking for builders to publish a real agent and tell me where it breaks — not vanity signups.
Try it: chat.indie.money · I'm the founder, AMA in the comments.
18 Comments
This is one of the more interesting agent marketplace writeups because you are treating per-run economics as first-class instead of bolting payments on later.
The part I would pressure-test hardest is buyer trust. If a buyer is paying per run, they eventually need a clean record of: what was requested, who paid, cost basis vs markup, whether the run succeeded, what happens on retry, and when a refund is possible.
I am working on SettleMesh from an adjacent angle — public access, login, usage billing, checkout, and MCP install flow for agent-built apps — and the recurring lesson is similar: once agents can trigger paid work, the backend starts to look like a ledger. If the product makes that run ledger visible to both builders and buyers, it becomes more than a marketplace; it becomes a trust layer.
"the backend starts to look like a ledger" is basically the founding observation, one of our internal lines is "blockchain for accounting, not theater". thats the whole reason settlement is on-chain here: the run record IS the product, not a reporting feature bolted on top.
Where we are against your checklist: what ran, who paid, what was authorized vs actually charged, and refund behavior all live in the run history today, every run leaves a signed receipt. where youre ahead of me: cost basis vs markup transparency. right now a caller sees the price and the authorized max, not the builder's underlying cost split, and im honestly still torn on how much of that to expose, there's a tension between buyer trust and builder margin privacy. curious how you're handling that in settlemesh, and more generally how the MCP install flow deals with a buyer trusting an app it installed programmatically. feels like our problems rhyme.
the history-as-moat instinct is right, but it only holds if the history is expensive to forge. a signed receipt proves a run happened — it doesn't prove it was a real buyer getting real value vs. the seller (or a botnet) padding the count. the moment "ran 4,000 times for 60 buyers" becomes the pitch, it becomes the thing people game. so the moat isn't the receipts, it's whatever makes them costly to fake: verified distinct payers, outcomes buyers actually confirmed, disputes/refunds attached to the same ledger. a clean run count with no downside signal is a vanity metric a clone can manufacture faster than you can. if the beta is really testing whether history is a moat, the thing to test is whether a motivated seller can inflate their own — because that's exactly what they'll try the day it starts driving sales.
you're right, and this is the second time someone sharp has landed on exactly this point, so its clearly the real one. a raw run count is a vanity metric a clone forges faster than i can earn it, agreed. the direction im convinced by is your list: weight history by whats costly to fake, returns from DISTINCT payers with stake at risk, outcomes the buyer confirmed, disputes and refunds on the same ledger so a bad run is as visible as a good one. self-dealing then costs N independent staked identities, not one botnet. not shipped yet. and youre right the beta should test whether a motivated seller can inflate their own, thats the adversarial case not the happy path. if you want to keep poking at the design, genuinely here for it.
honestly this has been the most useful thread i've been in here — you're building the thing i keep poking at from the security side. at some point these stop being forum-sized questions: sybil economics, stake-scaling, slashable reputation, the semantic residue left after structural checks — that's a build, not a comment. this is the kind of system i actually work on. if it'd be useful to take it deeper than a thread — pressure-test the trust layer properly, or help build it — i'm up for it. either way i'll keep poking at the public version, it's a genuinely good problem.
The copyability question fez33 raised is the one I'd obsess over, because it's what actually decides whether this is a marketplace or a directory.
I run a marketplace on the publishing side (Automateed, an AI book creator where authors publish and keep 85%, no exclusivity), so different vertical, same physics. The lesson that took me too long: the listed artifact is never the moat. Anyone can rebuild a workflow, and any author can take their book elsewhere. What people don't copy is demand. Whoever owns the buyers owns the marketplace, and the creator-friendly split is really just a supply-acquisition tactic to get enough listings that buyers show up in the first place.
So I'd reframe your own question. The per-run track record isn't the moat by itself either; it's the mechanism that lets a buyer trust a listing enough to pay without a human in the loop. That trust is what compounds and what's genuinely hard to copy. The 100% split gets you builders; the trust layer plus demand is what stops them (and their cloners) from routing around you.
Betting on demand-side liquidity over artifact lock-in is the slower build, but it's the one that survives the day a competitor copies your whole catalog.
this is the most useful comment i've gotten all launch, thank you. and yeah, i'd sign every line of it: the split is supply acquisition, the artifact is copyable, demand is the moat. my bet is narrower than "track record wins": it's that in an agent market the buyer often ISNT a human reading reviews, its another agent or a no-code reseller, and neither can evaluate a listing by vibes. so the trust mechanism has to be machine-readable (receipts, run history, refund behavior) or the demand side cant form at all.
curious from the automateed side: what actually moved demand for you early, before you had liquidity? thats the part i'm staring at now.
The per-run economics insight is interesting - most marketplaces make builders think in terms of one-time transactions because that's what their payment model rewards. Here you're inverting that by making recurring revenue visible and ownable. The 10% cash-out fee vs 0% ongoing is a clever nudge toward thinking in terms of assets. Question: does the resale layer (Producers) become the main revenue driver once builders realize they've built something worth $X in repeat revenue? That would flip the default unit of value from "agent" to "portfolio".
that flip is basically the thesis. if a builder looks at their dashboard and sees "this agent produced $340 in repeat revenue this quarter", the mental unit stops being "template i sold once" and becomes "asset i hold". resale is the natural next step from there, both directions: builders licensing out, and Producers assembling portfolios of other people's agents with their own markup.
whether resale becomes the MAIN driver, too early to say honestly. right now the whole game is getting the first builders to a first paid run so the asset framing stops being a pitch and starts being a screenshot.
The "100% of your price" framing is the interesting part — most agent marketplaces I've seen take a cut or gate on their own payment rail. The copyability question feels separate from the ownership one already raised here though: once an n8n-compatible agent is proven and its price is public, what stops a buyer from just rebuilding the workflow and listing it themselves? Curious whether the per-run track record is the actual moat, or if it's the resale/Producer layer doing that work.
honest answer: nothing stops them rebuilding the workflow. the JSON was never the moat and i wont pretend it is. what they cant rebuild is the published agent's history: every run is a signed receipt, so "this exact agent ran 4,000 times for 60 distinct buyers" lives with MY listing, not the clone. a fresh copy starts at zero trust with a public price anchor above it, which is a bad place to sell from.
also worth saying: the buyers i care about most cant rebuild a workflow at all. thats the Producer side, people who resell a working agent through a chat page without touching the canvas. for them the question isnt "could i rebuild this", its "does it work and who vouches for it".
is that enough of a moat? genuinely dont know yet, thats half of what the beta is for.
What stood out to me is that you're changing who owns the economics, not just where agents are sold.
The long-term question isn't whether builders can publish agents. It's whether they start thinking of them as owned assets that compound over time instead of one-off automations. That's a very different marketplace dynamic.
honestly you put it better than my post did. the publish-and-charge part stopped being interesting the day apify turned it on for 20k tools ... fine, that race is over.
the thing i actually care about is whether builders start treating an agent like something they own. right now most of us price like freelancers, me included for years ... do the work, invoice, done. the agent kept running, someone else kept the upside.
no idea yet if people make that shift tbh. the run history + resale stuff is built so the asset framing is real, but a data model doesn't change how people think about their own work. that's kind of what the beta is for ... finding the few who already think that way.
That's exactly what made me think there's a bigger strategic decision underneath your beta.
I don't think it's really about whether builders adopt the asset mindset. It's about a business decision that becomes much more significant if they do, and I don't think I can explain the reasoning properly in a thread without oversimplifying it.
If you're interested, what's the best email to reach you on?
appreciate that! two things: i try to keep the thinking public where i can, so honestly even the oversimplified version here would interest me (and probably others reading). if it genuinely needs the long form, my X DMs are open: @pvdyck1
Happy to share the short version.
The thing I'd be watching isn't whether builders start thinking of agents as assets.
It's what that shift quietly commits the marketplace itself to becoming.
I don't think the consequences of that decision are obvious at first, and I'd rather not pretend I can do the reasoning justice in a few sentences.
That's the part I found interesting.
Yar aryan bahi ye btana k coment outreah se business ata ha aur clint convert hote han ya ni ap bohat sare coment krte han har post par meri trah
Sometimes, yes.
The main goal isn't immediate clients—it's starting good conversations with founders. The business comes from a small number of those conversations, not from the comments themselves.