We're building open-source financial infrastructure

opensource.finance is an open-source project building self-hosted infrastructure for fraud and AML transaction monitoring. Instead of renting a closed vendor system, fintechs and payment companies run the engine on their own servers and can read every rule it applies. Code is Apache 2.0; the project is published by Reaktif LLC.

Osprey (live, v0.1.0): the transaction monitoring engine. Osprey Narrator (v0.1): a fine-tuned model that drafts CEL rules and SAR narratives, on Ollama and HuggingFace. Studio (building): a visual rule builder. Cases (roadmap): an alert triage workspace. Adapters (roadmap): ISO 20022, gRPC and GraphQL ingestion; Osprey accepts JSON over REST today.

Osprey is one Go binary that scores each transaction against CEL rules and returns an alert decision, a risk score and the reasons. It ships with 12 FATF-inspired starter rules, and a rule change is live without a restart. Measured on a laptop with v0.1.0: about 1 ms median per request, and 380 evaluations per second at p99 66 ms with 10 concurrent clients.

Tazama, the open-source monitor Osprey's founder helped build, runs as 7+ microservices on Kubernetes and needs a platform team. Osprey packages the core into a single binary one engineer can run. Enterprise vendors such as Sardine, Unit21 and ComplyAdvantage charge roughly $30,000 to $740,000+ a year and hold your data; Osprey is Apache 2.0 and runs inside your own infrastructure.

Osprey Narrator is a fine-tuned language model (v0.1). It drafts CEL rules from a plain-English description of a fraud pattern, and drafts SAR-style narratives from an Osprey alert. It runs locally through Ollama or HuggingFace. Review its output before you use it.

Yes. The engine is Apache 2.0 and the Narrator model is published openly on HuggingFace and Ollama. You run them on your own hardware or cloud with no per-seat or per-transaction fees, and transaction data never leaves your infrastructure.