Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

Ojuri

ojuri.io

  • Free plan
  • Open source
  • Windows
  • Linux
  • Web
  • API
  • Self-hosted
8.0 Club score Visit ojuri.io

What Ojuri is

Ojuri is a self-hosted fraud detection engine that scores transactions and returns accept, review, or decline decisions. It is aimed at fintechs, payment gateways, and mobile-money platforms. Applications can submit transactions through the POST /v1/predict API, and a Node SDK supports typed requests, retries, and webhook signature verification. Real-time scoring uses an XGBoost model served through ONNX Runtime; another agent tracks account, device, and payee relationships to identify fraud rings and mule networks. Decisions include reason codes and the data considered by the model in an audit log. Sentinel provides a review queue, decision stream, model registry, and analyst overrides that can become training data. Learning tools monitor drift, retrain on verified labels, and gate model promotion on evaluation checks. A separate investigation agent can prepare reports for declined transactions using an optional self-hosted language model. Ojuri says transaction data stays within the operator’s infrastructure. It is MIT licensed, with no per-call fees or rate limits. Setup uses Docker Compose and requires Docker and Node 20.

Who it suits

Ojuri is intended for fintechs, payment gateways, and mobile-money platforms that want to run fraud detection in their own infrastructure. It may suit teams able to deploy and operate a Docker Compose setup.

Good to know, and what to watch

  • Self-hosted transaction processing
  • Returns accept, review, or decline decisions
  • Audit log records reasons and model input data
  • MIT licensed with no per-call fees or rate limits
  • Quickstart requires Docker and Node 20
  • Investigation model is optional and downloads about 7.6 GB of weights
  • Self-hosting is not a compliance certification

Free

Free

Ojuri

Free

Free plan

Getting started

  1. Open the Ojuri website
  2. Prepare Docker and Node 20
  3. Start the quickstart with Docker Compose
  4. Follow the separate Windows setup when using Windows
  5. Call the POST /v1/predict API or use the Node SDK
  6. Enable the optional investigation model if needed

Limits to check

The free Ojuri plan is MIT licensed, costs nothing, and has no per-call fees or rate limits. The optional investigation model requires about 7.6 GB of downloaded weights.

The Club's review

Ojuri is a self-hosted fraud engine for fintechs, payment gateways, and mobile-money operators. It suits teams that want to run scoring, review, and model operations inside their own infrastructure. The trade-off is that deployment, capacity, and ongoing maintenance remain the operator’s responsibility.

Overview

Ojuri evaluates transactions and returns accept, review, or decline decisions through an omnichannel API. Its architecture combines live payment scoring with relationship analysis, analyst workflows, investigations, and model management rather than treating fraud detection as a single rules endpoint.

Keeping services in the operator’s infrastructure means transaction data stays within that boundary for scoring, training, and investigation. That can support data-residency requirements such as GDPR, NDPR, and CBN, but Ojuri is not a compliance certification. The quickstart requires Docker and Node 20; enabling investigation reports also downloads roughly 7.6 GB of model weights, so a lightweight deployment is not guaranteed.

Key features

Real-time decisions and fraud rings

The Real-Time Detection Agent scores payments on the authorization path using an XGBoost model served through ONNX Runtime. This is the part to consider when a decision must fit directly into payment authorization. The Pattern Analysis Agent follows account, device, and payee relationships over time, adding a way to expose fraud rings and mule networks that isolated transaction checks can miss.

Explainability and analyst control

Ojuri records each decision with reason codes and the data seen by the model. That gives investigators an audit trail for understanding why a transaction was accepted, reviewed, or declined. Sentinel adds a review queue, decision stream, model registry, and analyst overrides; those overrides can become training data, linking day-to-day review work to later model improvement.

Learning and investigations

The Model Learning Agent monitors drift, retrains on verified labels, and gates promotion on evaluation checks. Teams that already maintain reliable labels can use this to make model changes more disciplined; teams without that process will still need to establish the operational practice around it. The Fraud Investigation Agent creates reports for declined transactions with a self-hosted language model on a separate path from live scoring, keeping report generation from being part of the authorization path while adding substantial model-storage requirements when enabled.

Integration and controls

Applications call POST /v1/predict. The Node SDK provides typed requests, retries, and webhook signature verification. API keys, dashboard accounts and roles, signed webhooks, and idempotency keys cover common integration and retry-safety needs. Custom rules, machine-learning models, and device fingerprinting are supported.

Pricing

PlanPriceWhat it includes
OjuriFreeMIT-licensed software, no per-call fees, and no rate limits

The free plan is unusually unrestricted on usage terms: there are no per-call charges or published rate limits. It fits organizations willing to operate open-source software themselves. The price does not remove the infrastructure cost, including Docker and Node 20 requirements and the optional investigation model’s approximately 7.6 GB download.

Platforms

Ojuri is offered as an API, self-hosted deployment, web dashboard, and Windows setup. Separate Windows instructions broaden its fit beyond Linux-first operations, while the self-hosted model remains central to data control.

Who it's for

Ojuri is aimed at fintechs, payment gateways, and mobile-money platforms that need transaction decisions across channels and want ownership of the fraud stack. It is a strong match for teams with deployment and model-operations capability, especially where keeping transaction data inside their environment matters. It is a weaker fit for organizations seeking a managed service with minimal infrastructure work.

Pros and cons

Pros

  • No usage fees or rate limits: the MIT-licensed plan avoids per-call billing.
  • Broad fraud workflow: live scoring, relationship analysis, analyst review, investigation reports, and model learning are covered in one system.
  • Traceable decisions: reason codes and model input data support audit and investigation work.
  • Data stays in the operator’s boundary: useful for residency-sensitive deployments, without presenting that as certification.

Cons

  • Operational burden: deployment requires Docker and Node 20, and self-hosting shifts upkeep to the customer.
  • Investigation storage cost: the optional language-model weights add about 7.6 GB to the setup.
  • Model operations are not automatic governance: drift monitoring and promotion gates still depend on verified labels and an internal review process.

Alternatives

Choose FraudLabs Pro instead when a freemium, hosted-style option with Android, iOS, web, and API access and a 500-query monthly free tier is a better fit.

IPQualityScore Fraud Prevention is preferable for a freemium service offering self-hosted access, 10 custom rules, and 1,000 lookups per month on its free plan.

Pick Stripe Radar for AI fraud prevention and card-testing protection included with standard Stripe payments pricing.

Jube is another free, self-hosted open-source option under AGPLv3.

Choose MaxMind minFraud when pay-as-you-go risk scoring with no commitment or minimums is more suitable.

SAS Visual Statistics fits organizations seeking enterprise-scale statistical modeling on SAS Viya.

Tazama offers free open-source software, with hosting, implementation, and support requiring resources or a contract.

Ravelin provides a customized platform tailored to business goals, risk appetite, and setup.

Verdict

Ojuri is best for technically capable fintech and payment teams that want an open-source, no-per-call-fee fraud workflow under their own control. Its main reason to choose it is the combination of live scoring, relationship analysis, explainable decisions, analyst review, and model-learning tools in a self-hosted system. Look elsewhere if you want a managed deployment or do not want to carry the infrastructure and model-operations workload.

The verdict

Ojuri combines transaction scoring with analyst review and model-learning workflows while keeping processing inside the operator’s infrastructure. Teams should account for deployment requirements and treat its data-residency support as distinct from compliance certification.

Questions people ask

What does Ojuri cost?

The Ojuri plan is free at no cost, with no per-call fees or rate limits.

Where does Ojuri run?

Ojuri runs self-hosted in the operator’s infrastructure. It supports API, web, Windows, and self-hosted use.

How can applications send transactions?

Applications can call the POST /v1/predict API. The Node SDK supports typed requests, retries, and webhook signature verification.

Does Ojuri provide explainable decisions?

Decisions include reason codes and the data seen by the model, recorded in an audit log.

Is Ojuri open source?

Ojuri is MIT licensed.

What the maker publishes

Analyst dashboard
Sentinel includes a review queue, decision stream, model registry, and analyst overrides that can become training data. source
Analyst workflow
Sentinel provides a review queue, live decision stream, rule editor, model registry, audit log, investigations, and user administration. source
Architecture
The system uses four agents: real-time detection, pattern analysis, model learning, and fraud investigation. source
Audit trail
Ojuri records each decision and the feature snapshot used at inference before returning the response. source
Benchmark result
The repository reports 34% fraud detection before retraining and 98.8% after one label-driven retraining round, with 1.1% false positives in its simulation. source
Compliance
Ojuri says self-hosting can help with data residency requirements including GDPR, NDPR, and CBN, and explicitly says it is not a compliance certification. source
Data handling
Ojuri says its services run in the operator’s infrastructure and transaction data does not leave that boundary for scoring, training, or investigation. source
Data residency
The maker states that scoring, training, and investigations run inside the customer’s infrastructure without transaction-data egress. source
Deployment
The quickstart uses Docker Compose and requires Docker and Node 20; the investigation model is optional and downloads about 7.6 GB of weights when enabled. source
Explainability
Decisions are recorded with reason codes and the data the model saw, so their rationale can be traced in the audit log. source
Explainability
Responses include reason codes, feature contributions, observed values, model version, threshold, and audit ID. source
Extensibility
Custom pre-model and post-model rules can be edited without restarting and normally update within 30 seconds. source
Feature extensibility
The default model exposes 64 features and supports defining additional signals without code changes. source
FIA resource limit
The optional investigation service downloads about 7.6 GB of model weights and needs about 16 GB of free RAM. source
Fraud-ring detection
Pattern analysis identifies shared-device clusters, relationship groups, and emerging transaction rings. source
Fraud rings
The Pattern Analysis Agent tracks account, device, and payee relationships over time to identify fraud rings and mule networks. source
Integrations
Applications can call the POST /v1/predict API, and the Node SDK supports typed requests, retries, and webhook signature verification. source
Integrations
The stack uses Kafka, Redis, PostgreSQL, ONNX Runtime, Docker Compose, NGINX, and Prometheus/Grafana components. source
Investigations
The Fraud Investigation Agent creates reports for declined transactions using a self-hosted language model and runs on a separate path from live scoring. source
Latest release
The homepage changelog identifies v1.7.0, dated September 28, 2026, as the current release shown there. source
Launch date
The project’s comparison page identifies v1.0 as released on June 7, 2026. source
Learning loop
Reviewer overrides, chargebacks, and disputes can become labels for later retraining. source
License
The repository is released under the MIT license. source
Limit: hosted offering
The project does not provide a hosted SaaS deployment or per-transaction service; customers operate the infrastructure themselves. source

Checked against the maker's pages on 15 Oct 2026.