Launched this week
AI agents are making real decisions — executing trades, accessing patient data, filing documents. But there's no way to prove what an agent is, what it's allowed to do, or what it actually did. Kakunin fixes that. It issues X.509 certificates to AI agents, monitors their behavior in real time, and auto-revokes access when risk crosses a threshold. Regulator-ready audit reports in one click. Works with LangChain, CrewAI, AutoGen, CAMEL, Mastra, Vercel AI SDK. Python and TypeScript SDKs.
Interactive
Free Options
Launch Team / Built With
Retainly Email Content Spam Checker
Does it delay the transactions, and if yes, then by how much ?
@raajesh_rp Nice question. Kakunin adds zero latency to agent transactions.
Architecture is out-of-band, not inline:
Cert issuance — happens once at agent registration. KMS signs (~200–400ms). Never repeated per transaction.
Behavioral events — your agent POSTs to /api/v1/agents/{id}/events fire-and-forget. Kakunin queues via QStash and returns 202 immediately. Event processing (risk scoring, anomaly detection) runs async in background.
Cert verification — standard X.509 check. Done by the verifier locally against the cert PEM. No Kakunin server call required.
Kakunin is a compliance observer, not a transaction proxy. It never sits in your agent's critical path.
Only latency touchpoint: if you build a gate that blocks an agent when cert is revoked — that's your architecture choice, not Kakunin's default.
How difficult is Kakunin to integrate with our existing AI Trading agent? How many days will it take roughly?
@allwinbose Good one — trading agent is actually ideal fit (MiCA covers crypto asset markets directly).
Integration is 3 steps, ~1–3 days:
Day 1 — Agent registration (2–4 hrs)
{ "name": "trading-bot-v2", "description": "..." }
Returns agent ID + X.509 cert. Store both. Done.
Day 1-2 — Event ingestion (4–8 hrs)
Fire-and-forget POST after each meaningful action:
{ "action_type": "trade_executed", "metadata": { "pair": "BTC/EUR", "size": 1.2 } }
No blocking. No await. Wrap in try/catch, log warning on failure, never throw.
Day 2-3 — Webhooks for alerts (optional, 2–4 hrs)
Register webhook → Kakunin pushes risk.alert when score ≥ 0.75. Trading agent can pause itself or alert ops team.
What they don't need to build: risk scoring, anomaly detection, compliance report generation, cert lifecycle — all Kakunin-side.
SDK available (npm i @kakunin/sdk) cuts Day 1 to ~1 hr.
What's pricing — per agent, per event, or flat?
@ravinath_mahto Very reasonable pricing Ravi, you can check here, its per agent, and the 2 tiers have different quotas. https://www.kakunin.ai/pricing
How do you handle high-frequency agents? (trading bots firing 1000s of events/min)
@innocent_ritesh Built for it. Event ingestion is async and non-blocking — your agent fires and forgets, we queue via QStash and process in the background. No per-event latency hit.
For true high-frequency (1000s/min), we support event batching so you can bundle N events in a single POST and reduce network overhead. Risk scoring runs on sliding windows, not per-event — so a 10k event burst doesn't trigger 10k scoring jobs.
Rate limits are generous for trading workloads and configurable per tenant. Hit us up if you need custom limits for your volume.
What happens if an agent's cert is compromised — how fast is revocation?
@anshu_kumari9 For auto-revocation: if an agent's risk score hits 0.85+ (our high-risk threshold), revocation triggers automatically — no human needed. - more detailed response provided in comments on a similar query.
mailX by mailwarm
Agentic KYC sounds cool. Curious to see how this plays out
@daniel_nwankwo Thanks Daniel! We're seeing real demand from fintechs and trading platforms needing to prove their AI agents are compliant before regulators come knocking. MiCA enforcement kicks in hard this year — timing feels right. Happy to share early access if you want to kick the tires.