Platform Architecture

How Automize is built – a cloud control plane, distributed execution, and an audited, tenant-isolated data model. Written for architects and technical evaluators.

Two Planes: Orchestrate in the Cloud, Execute at the Edge

Automize separates the control plane – the cloud application that designs, schedules, and governs automation – from the data plane, where bots run the work inside your own environment. Bots dial outbound only: they poll the control plane for commands and stream results back, so there are no inbound ports to open on your network.

Automize topology: users and the cloud control plane connect over outbound HTTPS to bots running in the customer environment, which reach external systems. Cloud API flows (no bot required) USERS Builders & administrators Operators run & monitor CONTROL PLANE cloud · automize.co.za Web UI & REST API Orchestrator & Scheduler Queues & Pools Identity · RBAC · Audit Logical tenant isolation Outbound HTTPS only results commands DATA PLANE your environment Attended bot desktop · human-triggered Unattended bots servers · scheduled EXTERNAL SYSTEMS ERP / Finance HRIS / Payroll REST & SOAP APIs Web & legacy UIs HTTPS
Bots initiate every connection outbound over HTTPS. The control plane never reaches into your network; it hands out commands only when a bot asks for them.

See the Trust Center for hosting, data residency & sub-processors

The Life of a Run

Every automation follows the same governed path from design to a tamper-evident record. Failures do not restart from zero – typed retries and resume-from-step pick up where the run left off.

Run lifecycle: build, enqueue, dispatch, execute, settle, with a retry and resume-from-step loop. 1 Build Design a process or cloud flow 2 Enqueue Scheduled or triggered into a pool 3 Dispatch Bot heartbeat pulls the command 4 Execute Attended, unattended, or in-cloud 5 Settle Result appended to the audit chain typed retry · resume from step

Five Ways a Run Starts

Automation is event-driven. The same process can be started by any of these, alone or in combination.

Schedule

Cron-based schedules run a process at a set time or interval, without anyone lifting a finger.

Webhook

An external system starts a run by calling a per-process URL – tokenised and optionally HMAC-signed to prove integrity.

REST API

Start, queue, and resume runs programmatically – post to a process to run it on demand from your own systems.

Event

Platform events fire a run in response to what happens elsewhere in your workspace, with priority ordering.

Manual

Operators start attended or unattended runs directly from the app whenever they need to.

Three Ways to Execute

The same process definition can run in whichever place fits the work – no rewrite required.

Attended

A person triggers the bot on their own desktop – ideal for human-in-the-loop steps, approvals, and tasks that need judgement alongside automation.

Unattended

Bots run on servers to a schedule or trigger, with no operator present. Pools distribute work across bots and apply back-pressure when demand spikes.

Cloud flows

API-to-API integrations run entirely in the control plane – no bot, no desktop. Fan-out, join, and mapping steps orchestrate data between systems directly.

Deployment Posture

The control plane is fully managed for you. Execution happens wherever your systems and policies require.

Managed cloud

The control plane runs as a managed service – nothing for you to provision, patch, or scale. There is no infrastructure to stand up before you begin.

Bots in your environment

Bots run on your own desktops and servers, reaching internal systems locally. They connect out over HTTPS on a strict egress allow-list – no inbound ports, no wildcard outbound.

On-premises option

Where regulation or policy requires it, the web portal can be deployed on-premises as a scoped engagement, sized to your environment.

API-First Integration

Automize connects to systems through their APIs wherever one exists, with a shared authentication layer that speaks the common enterprise protocols. UI automation is the fallback – used only when there is no API – and it is hardened with self-healing selectors that adapt when a screen changes.

Integration layers: Automize execution passes through a connector authentication layer to reach external systems. Automize execution · processes, cloud flows & bots CONNECTOR LAYER · AUTHENTICATION OAuth 2.0 PKCE OAuth 1.0a API key HMAC AWS SigV4 Basic ERP HRIS / Payroll Finance / Ledgers REST / JSON SOAP / Legacy Web UIs
One authentication layer, many systems. Credentials are stored encrypted and resolved at run time – never embedded in a process definition.

Built to Be Extended

When the built-in library does not cover something, your team can extend the platform without waiting on ours.

REST API

A documented REST API (OpenAPI 3.1) covers runs, processes, and credentials, with a live collection you can import and try.

Plugin SDK

Add custom actions and connectors in Python – subclass a plugin base, declare a JSON schema, and the designer renders the inputs automatically.

Connector catalogue

A large built-in catalogue, plus a generic HTTP connector for any REST or SOAP endpoint with per-tenant host and header configuration.

Template marketplace

Install ready-made process templates from the marketplace and adapt them to your own workflow.

AI That Assists, Not Replaces

AI speeds up building and running automation while you stay in control. It is powered by OpenAI and Anthropic, with your choice of provider and model per workspace.

In-app assistants

Conversational assistants help you build processes, answer product questions, and work alongside your wikis.

Document understanding

Vision-capable models read invoices, forms, and other documents and images, extracting structured data your automation can act on.

Authoring copilots

AI suggests field mappings and groupings, renames steps, diagnoses failures, and drafts flows – every suggestion is yours to accept or reject.

Your choice of model

Choose OpenAI or Anthropic per workspace, and pick the model per task. No single provider is locked in.

Secrets Live in a Vault, Not in a Process

Credentials are never written into an automation. They are held in an encrypted vault and resolved by reference only at the moment a bot needs them.

Encrypted vault

Credentials are encrypted at rest. Names and usernames are browsable for reuse; the secret itself is never displayed in the interface.

Your choice of backend

Store secrets in the platform vault, AWS Secrets Manager, or the bot host operating-system keychain. The database only ever holds a reference.

Reference, never embedded

Automations reference a credential by identity. A scanner flags any hardcoded secret, so plaintext cannot creep into a definition.

Governed access

Per-credential grants (metadata-only, use-at-runtime, or edit), rotation and expiry policies, and a hash-chained access log.

Defence in Depth

A request crosses three layers before it can touch data. Each layer is independent – a gap in one is caught by the next.

Three security layers: perimeter, identity and access, then encrypted data with a tamper-evident audit chain. PERIMETER TLS 1.2+ HTTPS-only CSRF tokens Rate limiting CSP IDENTITY & ACCESS SSO (SAML / OIDC) RBAC ShareGrants Tenant isolation Least privilege DATA Encrypted at rest Private DB tier Tamper-evident audit chain PII redaction request ↓
Sensitive writes are recorded in a hash-chained audit log – any tampering with a past entry breaks the chain and is detectable.

Full security controls, compliance & frameworks in the Trust Center

Reliability & Scale

Automation that behaves predictably under load and recovers cleanly when something breaks.

Pools & queues

Work is distributed across bots through pools, with back-pressure so a demand spike queues rather than overwhelms.

Typed retries

Failures are classified, so a transient network blip retries automatically while a genuine business error surfaces for a human.

Resume from step

A run that fails part-way does not start over – it resumes from the step it reached, so no completed work is repeated.

SLA monitoring

Runs and pools report against service-level targets, so a stalled queue is visible before it becomes a missed deadline.

Governance & Lifecycle

The controls that keep automation auditable and deployments safe as the platform evolves.

Tamper-evident audit

Sensitive operations are hash-chained end to end. The record is append-only and verifiable – you can prove what happened and when.

Safe deployments

Schema changes and feature flags move through continuous integration with automated checks before anything reaches production.

Observability

Runs, queues, and integrations emit structured logs and metrics, so operators can trace a single run from trigger to result.

Compliance & Data Governance

The controls a security or compliance reviewer looks for. The full detail – hosting, sub-processors, and frameworks – lives in the Trust Center.

EU data residency

Customer data is hosted in the European Union. Egress to other regions is denied at the network layer, not just by policy.

Encryption everywhere

TLS in transit and encryption at rest across the database, file storage, and backups – no plain-text tier.

Tamper-evident audit

Sensitive actions are chained with HMAC-SHA-256 and verified nightly. Audit exports are written to write-once (WORM) storage.

Erasure & offboarding

A tenant-erasure process removes customer data on offboarding, while preserving legally required audit records under lock.

PII-safe logging

Logs are scrubbed of personal data and secrets before they are written to disk or sent to error tracking.

Encrypted backups

Automated, encrypted backups guard against data loss, retained on a defined schedule.

Hosting, sub-processors & compliance frameworks in the Trust Center

Want the deeper technical detail?

Bring your architecture and security teams. We are happy to walk through the topology, the security model, and how Automize fits your environment.