Block-based workflow editor
Prompting, testing, versioning, multi-prompt, cost control
Resume line
Retrieval Augmented Generation (RAG)
Roles and permissions
Security, data transparency
Web
Data Transformer
Decision Maker
Extract Maker (with HTML)
Key / Knowledge Manager
Web Creator
SaaS Email, Integration
Cloud privacy managed
On premise
PLATFORM · ARIFLOW
Build and control AI automations and features without reinventing your infrastructure.
Visual editor, automated tests, real-time cost monitoring. From prototype to production.
ariflow: workflow editor
WHAT IT IS
The low-code platform for orchestrating generative AI inside your processes.
ARIflow is a visual editor that composes LLM calls, document search, web actions, conditional decisions and data transformations into a single flow. A flow that then goes to production with an API endpoint, measurable quality gates and costs tracked per invocation.
Hybrid AI + code system
Every node can be an LLM call, a deterministic rule, a SQL query or a custom block. You combine AI flexibility with classical-logic predictability.
Built-in quality gates
Checkers that validate specific cases, evaluators that aggregate metrics over time. AI regressions become numbers, not hunches.
End-to-end data control
Deploy in managed cloud, in your Kubernetes cluster or fully air-gapped. Encrypted secrets, isolated tenants, self-hosted models available.
WHO IT'S FOR
For teams that need to ship AI to production and keep it under control.
ARIflow gives you a single environment to build, test and ship your AI solutions to production: with quality and costs under control, from the first prototype to everyday use.
01
Ideal for:
CTO & Head of Engineering
Standardize how the team builds AI, without reinventing the process for each project.
02
Ideal for:
AI Engineer / Solution Architect
Prototype fast, measure quality, deploy with endpoints and auto-generated documentation.
03
Ideal for:
IT Manager & DevOps
Know where data runs, control LLM costs, have audit trails on every run.
04
Ideal for:
Software House & System Integrator
Reuse workflows across clients, keep tenants isolated, resell at your own margins.
What it does, concretely
Automate heterogeneous documents at scale
Conversational assistants with RAG on your documents
Agentic decisioning on tickets, leads, requests
Automate tasks that demand high precision
Automate boring, repetitive tasks
Let AI work with your data and documents, wherever they are
HOW IT WORKS
Five moments, one continuous flow.
Every workflow goes from idea to production following the same cycle: different tools for each phase, uniform discipline throughout.
01
Design
Define inputs, expected outputs and success metrics.
02
Build
Connect nodes and blocks. Every run is tracked: you see input, output, cost and latency of each single node.
03
Validate
Define checkers on specific cases and evaluators on aggregate metrics. Compare versions in the ResearchTree.
04
Publish
Publish the stable version. ARIflow generates HTTPS endpoints and API documentation automatically.
05
Monitor
Real-time dashboard: LLM costs per workflow, latencies, checker success rates, quality drift.
WHY ARIFLOW
Four things no other AI platform brings together.
AI workflows that test themselves
Checkers on specific cases + evaluators on aggregate metrics. Quantitative comparison between prompts, versions and models, not "we tried something", but "we know which is better, by how much, on what dimension".
Your data never leaves your perimeter
European cloud, your Kubernetes cluster or fully air-gapped. With ARIrun even LLMs can run inside your house. Zero data sent to OpenAI, Anthropic or Google if you choose so.
Pay for what AI does, not for who's logged in
Real-time consumption tracked per project, workflow, version, API token, individual model. The Price List is configurable per workspace: you can resell at your own rates.
ARIanna is built on ARIflow
ARIanna, the platform's built-in assistant, is itself built with ARIflow. If it holds up our own product in production, it holds up yours too.
INSIDE ARIFLOW
Everything you need, already inside.
Not a generic toolbox but a curated library, organized by functional families. Covers reasoning, retrieval, document ingestion, web actions, control flow, conversation, storage and custom code, without adding external vendors piece by piece.
Library of blocks and nodes
36 standard nodes plus every sub-workflow you build. Every workflow becomes a reusable block: your library grows with you.
System & Flow Nodes
Control flows, conditions, iterations and data transformations.
External Integrations
Interact with external systems and services (API, messaging, cloud).
Database Interactions
Read, write and query structured and indexed data.
Conversational Interactions
Read and monitor the state of interactions.
AI Interaction
Use LLM models to generate and analyze content.
Document Interactions
Handle unstructured files and content like PDFs, images and text: analysis, transformation and document preparation.
Web Actions
Interact directly with web pages.
Two-level automated testing
Checkers that answer "did this specific case produce the right output?", with field-by-field correctness, not just pass/fail. Evaluators that answer "how good is this version on average?", with comparable aggregate metrics.
✓
Field-by-field correctness, not just final output
✓
Evaluators comparing multiple LLM models on the same case
✓
Tests re-run automatically at every publish
✓
ResearchTree to compare versions and branches
Real-time cost monitoring
Every LLM call, every vector query, every document page: tracked against the workflow that generated it. Multi-dimensional dashboard: per project, version, API token, individual model.
✓
Breakdown per single run and per aggregate
✓
Price List configurable per workspace
✓
Consumption limits per API token with enforcement
✓
Cost attribution per integration
Versioning + one-click deploy
Publish freezes a version. Deploy assigns an HTTPS endpoint with automatically-generated API documentation. Previous versions stay callable: friction-free roll-forward in production.
Roles, permissions, secrets management
Four native roles (Maintainer, Developer, Tester, Executor) plus fine-grained per-resource roles. API keys encrypted at rest, never exposed in logs, referenced by name inside workflows.
Built-in observability
Every run is a permanent, inspectable object. See input and output of each node, latency, errors. Runs persist with the workspace: you can investigate AI behavior months later.
Native MCP, both directions
ARIflow is an MCP client: your workflows call external MCP servers. And it's also an MCP server: external AI agents can drive ARIflow in natural language.
WHERE IT RUNS
Three ways to install ARIflow. Same engine, your perimeter.
Change the infrastructure backbone, workflow logic stays the same. Go from managed cloud to full air-gapped without rewriting anything.
Fastest time-to-value
Managed cloud (EU SaaS)
Zero infrastructure to manage. Datacenters in the European Union. Multi-tenant isolated on Kubernetes. Perfect to start and scale without thinking about ops.
Guaranteed data residency
Customer-hosted (Kubernetes)
Deploy in your Kubernetes cluster: cloud or on-premise. Kafka or BullMQ as the backbone. Data never leaves your infrastructure, updates stay managed.
For regulated industries and defense
Air-gapped
Fully disconnected from the public network. Workflows pre-encrypted at build time, decrypted only inside your environment. With ARIrun even LLMs run self-hosted: no calls to the outside.
Let's see together if ARIflow is what you need.
Tell us about your use case: we'll show you the editor, testing and monitoring, and discuss pricing, deployment and roadmap in detail.
Cookies and tracking
We use technical cookies and — with your consent — analytics and marketing cookies to measure traffic and improve your experience. You can accept, refuse or change your choice at any time.
More info in the Cookies Policy →