Everything between the canvas and production
Draw it once. It compiles to a Java service, ships to your server, and answers requests - versioned, health-checked, reversible.
Field mapping
AI mapIncoming request
Ledger entry
Every workflow is three layers
Whether a flow moves one field or runs a whole settlement, it is described the same way: the systems it touches, how their fields line up, and the decisions in between.
Data sources
The first layer is everything the workflow touches.
You pick connectors from the library, configure them once, and drop them onto the canvas. An HTTP initiator gives the flow its endpoint; databases, brokers, file drops and partner APIs hang off it as nodes you can rewire by dragging an edge.
- Connectors are configured once and shared by everyone on the team.
- An HTTP initiator is what turns a flow into a callable endpoint.
- Change where the data comes from by rewiring a node, not the flow.
HTTP initiator
POST /v1/payments
PostgreSQL
customers.find_by_ref
Kafka
payments.events
Partner REST API
POST /settlements
Attribute mappings
The second layer is how the fields line up.
Drag a field from the request on the left to the field it becomes on the right. The builder matches what it can on its own and hands the awkward remainder to the AI mapper. The result is a mapping anyone can read at a glance - one line per field, nothing implied.
- customer.idparty_id
- customer.emailemailautoauto-matched
- amount.valueamount_minor
- Fields whose names and types already agree are matched for you.
- The AI mapper proposes the rest; you accept them or draw your own.
- Every mapping stays on screen - nothing is buried in generated code.
Logic
The third layer is the decisions the flow makes.
The Logic Builder carries 150+ operations across 13 categories - conditions, transformations, validation and the rest - assembled into a graph that reads like the process it describes. When you would rather read than click, the same logic renders as AXL text. The canvas and the text are the same thing, in two views.
value = substring(baseData, cursor, length)nextCursor = cursor + length
1,890+ connectors, from raw protocols to payment rails
A connector is a system your flows can call: a database, a broker, an API, a file drop. Configure it once for your team, then reuse it in every workflow you build.
- API protocols
REST · SOAP · GraphQL · gRPC · WebSocket · form and file endpoints
- Databases
PostgreSQL · Oracle · SQL Server · MySQL · MongoDB · JDBC-compatible engines
- Messaging & queues
Kafka · RabbitMQ · ActiveMQ · MQTT · JMS brokers
- Enterprise systems
SAP · Salesforce · Microsoft Dynamics · Oracle applications · in-house ERP
- Storage & files
SFTP · FTPS · S3-compatible object storage · Azure Blob · network shares
- Cache & in-memory
Redis · Hazelcast · Infinispan
- Agentic AI
OpenAI-compatible endpoints · self-hosted Ollama and vLLM · MCP tools
- Identity & security
OAuth 2 · JWT · LDAP · API key and request-signature schemes
Regional banking & payments
UAE · Pakistan · Bangladesh
Bank switches, wallets and settlement rails across the Gulf and South Asia - the integrations most platforms leave you to build from scratch.
Government & tax
Filing · e-invoicing · regulatory
Authority endpoints sit in the same library as everything else, so a regulated filing is drawn on the canvas like any other flow.
Connects with the systems you already run.
Logos identify systems Alphanetix connects to. They do not imply partnership, endorsement or customer status.
Functions your team owns, versions you can trace
Reusable logic lives in a shared library rather than being copied between flows - and every publish is a recorded, reversible event.
A library, not a copy-paste habit
User-defined functions are scoped to the team that owns them. Build one visually the same way you build a flow, or import logic your engineers already wrote in Java and give it a name the rest of the team can use. Either way it appears in the builder as one node.
Publishing is a commit
Versioning is git-backed. Publishing bumps a major or minor version, takes a commit message, and leaves an audit trail you can read months later - who changed what, when, and why. Rolling forward and rolling back are the same motion.
enrich_party
Imported from Java- v1.2
- v1.3
- v1.4
feat: handle partial settlement callbacks
Publishing v1.3 again is the rollback - same audit trail, no rebuild.
How a workflow reaches your server
Publishing is a sequence you can watch, not a handover to somebody else's cloud. The compiled service goes to a machine you own, and starts there.
Ship
The compiled JAR is copied to your server over SFTP.
payments-svc-1.4.jar
Start
Alphanetix opens an SSH session and starts the service.
ssh · java -jar
Health check
The platform polls the service until it answers healthy.
GET /health
Live
Traffic reaches your endpoint. Nothing left your infrastructure.
200 OK
Environments you draw
An environment is a set of your machines and the applications running on them. You arrange it on screen, then apply it.
- Where it runs
- Your own VMs or VPS - cloud, on-premises or a hybrid of both. Alphanetix ships software to your servers; it never becomes the place your data lives.
- Clustering
- The Environment Builder is a drag-and-drop map of your topology. Group applications into a local cluster on Hazelcast or a global one on Infinispan.
- Staged changes
- Edits to a topology stay a draft until you apply them, so the shape of production only changes when you say so.
- Access control
- Every environment carries its own IP allow and deny lists.
- Throughput
- Compiled, not interpreted, so there's no platform-imposed ceiling: throughput scales with the hardware underneath it, then with how many nodes you add.
production · your VPC
core · Hazelcast · local
- payments-svc
- ledger-svc
edge · Infinispan · global
- notify-svc
- audit-svc
An AI layer that already knows your platform
Generic assistants guess at your systems. This one is wired to the connectors, workflows and permissions you have already set up.
“When a payment webhook arrives, verify the signature, write the entry to Oracle and publish it to Kafka.”
- HTTP initiator
- Verify signature
- Oracle insert
- Kafka publish
Every node lands on the canvas bound to a connector your team has already configured - then you edit it like anything else you drew by hand.
Agents with a boundary
An AI agent gets a knowledge base to retrieve from and an explicit list of workflows it may call. Tools are default-deny: if you did not grant it, the agent cannot reach it. Guardrails bound what it is allowed to say and do.
The model runs where you decide - a local Ollama or vLLM server when the data cannot leave the building, or a cloud provider when it can.
Your platform, their assistant
Alphanetix ships an MCP server, so external AI assistants - Claude, Cursor, Copilot - can build, deploy and test workflows on your platform while bringing their own model keys.
The assistant does the drafting. The platform still does the compiling, the deploying and the health checks, under the same team permissions as a person.
See it on your own integrations
Bring a system you actually need to connect. We will build the flow with you and deploy it to a machine you control.