Deployment
Self-hosting isn’t a cut-down edition. SpeculaPro is one codebase: a self-hosted installation is the same software, running for a single institution.
The fastest way to go live. We run, monitor and update the platform; your team works alerts and cases from day one.
For banks and operators whose data can’t leave their environment. Customer data never reaches us.
Side by side
| SaaS | Self-hosted | |
|---|---|---|
| Who runs it | We do | Your team, on your infrastructure |
| Where the data lives | Our clusters, pinned to a region for data residency | Your data centre or private cloud; nothing has to leave |
| Tenancy | Many institutions, isolated by database row-level security | One institution, same software |
| Updates | Continuous | Versioned, signed releases you schedule; rule-pack updates signed by SpeculaPro |
| Packaging | Managed for you | Docker Compose for smaller installs; a Helm chart for Kubernetes |
| Notifications | Email, SMS and WhatsApp | Your own mail server; SMS if outbound traffic is allowed; WhatsApp through your own GOWA server |
| Internet access | Required to reach the service | Not required: the web app is built into the program |
What you run
SpeculaPro is a single Go program with the web app built in. There’s no separate Node.js server to operate. It runs on open infrastructure your team may already know.
One program, started with the roles you need: API, decisions, ingestion, stream processing, persistence, background work. Small installs run them all in one process; large ones scale each independently.
The system of record and the evidence store. In production we recommend Pigsty for high availability, connection pooling and backups.
Keeps every customer’s events in order and absorbs bursts such as salary days, so nothing is dropped under load.
Holds the velocity counters rules read in real time. Standalone, Sentinel or Cluster. It is a cache that can be rebuilt, never the source of truth.
Sizing
One SpeculaPro process running every role, a PostgreSQL primary with a replica, a small Redpanda cluster and Valkey with Sentinel.
Decision nodes for real-time traffic, stream workers up to the partition count, dedicated writers to the database, and Redis Cluster for the feature store. Engineered for 5,000 transactions per second sustained and 15,000 at peak, as targets.
Next step
Tell us about your volumes, your hosting rules and your team. We’ll recommend SaaS or self-hosting, and the sizing to start with.