Launching today
Sirenic
French & EU company data for AI agents, pay-per-call
8 followers
French & EU company data for AI agents, pay-per-call
8 followers
Official French & European company data as a pay-per-call API for AI agents. One call returns a KYB file: identity, officers, filed financials, BODACC alerts, and sanctions screening of the company and each officer vs the 5 official lists. Paid in USDC via x402, no account, no API key. Also an MCP server (25 tools).
Hi Product Hunt 👋 I'm Côme, solo builder from France.
AI agents can't sign up for anything: no email to verify, no card to enter, no API key to generate. So most data APIs stay closed to them. Sirenic flips that: the HTTP request carries its own payment.
You call an endpoint, get back a 402 with a price, your agent signs a USDC or EURC payment (x402, on Base), calls again, and gets the data. No account, no API key, and a failed request is never charged. Every paid response is Ed25519 signed, so an agent can verify offline that the data really came from Sirenic, unmodified.
It serves official French and European company data from open government registers: search 30M companies, full profiles, filed financials, BODACC legal alerts, one-call KYB files with sanctions screening of the company and each officer against 6 official lists, and a $1 intelligence report that cross-references every block into closed-list signals and a rule-based verdict. Belgium, UK, Norway, Estonia, Latvia and worldwide via GLEIF are in one unified schema too. Prices run $0.001 to $1.00 per call.
It's also an MCP server (29 tools), so Claude or Cursor plug in with one line:
It's early and I'm one person. I'd love feedback on the pay-per-call model, and which datasets would actually be worth a per-call price.
Happy to answer anything here 🙏
A bulk lookup endpoint would be huge, even if it costs more per call. Right now if an agent needs to screen a list of vendors or counterparties it has to loop one call at a time, which kills latency and adds unnecessary network chatter. Adding a single POST that accepts up to maybe 50 SIRENs and returns an array of KYB files in one response would make this way more useful for onboarding flows.
@yavuzotmanujpr
Good news that already exists! /v1/kyb/batch takes 2 to 100 companies in a single call (comma-separated SIRENs) and returns one KYB file per SIREN, plus nombre_demande/nombre_trouve to summarize the batch. Unknown SIRENs come back trouve=false. It's billed per company at 30% off the single-call price built exactly for list-based onboarding/screening. It's a GET today (the SIREN list fits in the URL); if you'd want a POST for very large lists, that's easy to add noted, thanks! Happy to share the schema.
The pay-per-call model is interesting because it changes how developers think about data access. Curious whether your biggest challenge has been improving the dataset itself, or making it easy for AI agents to know when they actually need to query it.
@aryan787544
Honestly, the dataset itself. Serving one clean KYB file means merging several official registers INSEE identity, INPI officers & accounts, BODACC events, sanctions lists keeping them fresh, and shaping any personal data to stay GDPR-safe (closed-list fields, no free text). That's been ~80% of the work, and it's the real moat. "When should an agent query" is more of a discovery problem: the MCP tool descriptions plus a $1 intelligence report that returns a plain verdict help the agent decide. But depth and quality of official data is the hard part.
@sirenic I like that distinction. It suggests the defensibility isn't really the API—it's the ongoing process of keeping the underlying data trustworthy.
I'm curious whether, over time, you see customers choosing Sirenic primarily because of coverage and freshness, or because they start trusting your judgment about which official data is actually worth querying in the first place.
@aryan787544
Both but in that order. Today people come for coverage + freshness; that's the entry ticket.
Over time, the durable part is the judgment: turning raw registry records into a verdict an agent can act on solid vs. fragile, an officer matching a sanctions list, whether the signals line up deterministic and explainable, not an LLM guess. Coverage can eventually be matched; that interpretation layer is harder to copy, and it's what agents actually want they don't need 40 raw fields, they need a decision.
The honest catch: you only earn trust in the judgment because the underlying data is fresh and official. So freshness isn't the moat it's the foundation the judgment compounds on. The $1 intelligence report is my early bet on that second layer.
@sirenic @sirenic I really like the way you've articulated that progression. Freshness earns the right to be trusted, but judgment is what compounds over time.
I've genuinely enjoyed this discussion. I have a few thoughts on how products like this can position themselves as AI agents become more capable. If you're open to it, what's the best email to reach you on?
paid a few cents in USDC and got back a clean KYB file with BODACC hits already flagged, way less friction than the usual vendor onboarding for a one-off compliance check.
@melisatekkol
Thank you :)
That's exactly the friction we're trying to kill a one-off compliance check shouldn't need a vendor account and a sales call. BODACC hits and sanctions are flagged inline so you get the verdict in a single response. Really glad it landed!