ARI is a private automation suite composed of two separate products: ARIflow, a workflow orchestration platform, and ARImodules, a set of pre-built vertical AI applications.
ARIflow is a low-code workflow orchestration platform. Users design automation workflows in a visual, node-based editor, connecting nodes that perform a specific operation (LLM prompts, document processing, database queries, API calls, browser automation, conditional logic, and iteration). Workflows can be tested automatically against defined evaluation parameters, including comparison of different LLM models to balance output quality and models' cost. Each execution is tracked, and its LLM cost can be verified both per individual run and projected across an expected volume of executions, allowing teams to estimate costs before deploying a workflow to production. Different versions and variants of a workflow can be branched and compared through a built-in versioning system (Research Tree), which tracks how each workflow evolves across iterations.
Once tested, workflows can be published as API endpoints with auto-generated documentation. ARIflow is aimed at technical users who build and maintain custom pipelines and solutions. It supports any LLM provider through an LLM-agnostic architecture, including self-hosted open-source models. ARIflow can be deployed as multi-tenant SaaS, as a single-tenant private cloud instance, or on-premise via Kubernetes on the customer's own infrastructure.
ARImodules are ready-to-use applications built on top of the ARIflow engine, aimed at operational users who needs to use specific AI applications without building workflows from scratch. Each module is licensed and priced separately, on a usage basis (not per user seat), with LLM inference costs included in the price.
ARIknow is a document question-answering module. Users upload documents into permission-scoped knowledge bases; queries are answered using hybrid retrieval (vector + BM25 keyword search), and every answer cites the source passage it was generated from. Knowledge bases are organized in parent-child hierarchies: granting access to a parent knowledge base automatically grants access to all of its child knowledge bases, without configuring permissions individually for each one.
ARIdb lets users query relational, NoSQL, or legacy databases in natural language. The model reads only the database schema, not the underlying data, and generates inspectable SQL queries; access is controlled per table and role. Results can be returned as a data table or rendered as a dashboard, in the visual format the user requests as part of the query.
ARIdoc extracts user-defined fields from documents (invoices, contracts, forms) regardless of layout. Each extracted field carries a confidence score; fields below a configurable threshold are routed to human review before being written to downstream systems (JSON, CSV, or direct API integration).
