Loading page
Platform
A conceptual overview for technical evaluators: how workflows are defined, how AI and rules combine, how executions run and how everything is observed.
Connected systems
CRM, ERP, inbox, data
Data
Events and records
AI
Classify, extract, enrich
Workflow
Rules and approvals
Result
Logged and observable
Architecture overview, not a feature checklist
System overview
Events enter through webhooks, schedules and the API. The workflow engine coordinates rules, AI and integrations, the execution engine runs each step, and every result is observable.
Hover a layer to lift it. The same architecture as a diagram is below.
Stores and versions workflow definitions as directed graphs of triggers, steps and branches.
Connectors share one contract: authentication, triggers, actions, pagination and rate-limit handling.
Executes classification, extraction, enrichment and recommendation steps that return schema-constrained output.
Evaluates deterministic conditions on structured data to route, gate and branch.
Runs workflow instances as isolated, queued executions with retries, timeouts and wait states.
Ingests events from webhooks, schedules and connectors, and emits execution events.
Aggregates run data into volume, success, duration and approval-latency views.
A management and trigger API mirrors what the product does in the UI.
Access control, secrets management, audit logging and tenant isolation span every layer.
Every execution produces a trace: steps, inputs, outputs, timing and errors.
AI + deterministic rules
An AI step produces structured output with a confidence value. From there, ordinary rules take over: thresholds, branches and approvals. The same input to a rule always gives the same result, and the AI step's output is logged so it can be reviewed.
Input
Record or document enters the step
AI step
Returns { segment, confidence } against a schema
Rule
IF confidence ≥ 0.80 THEN continue ELSE ask a person
Approval / action
Human decision or automatic action, recorded on the run