Platform Overview

This page is the single-stop summary of the Vextro platform for audit, procurement, and compliance review. It covers what the platform does, how data flows through it, what AI models power it, and what controls protect your data.

For deeper material on each topic, follow the links into the relevant sections.

What the platform does

Vextro orchestrates AI-powered workflows. Authors design flows visually, compose them from AI model calls, knowledge retrieval steps, and deterministic logic, and then run those flows from the UI, on a schedule, or through the platform API.

Concretely, the platform delivers four user-facing areas:

• Workflows - a visual builder where flows are composed and executed.
• Knowledge base - document collections indexed for retrieval- augmented generation.
• Models - the catalogue of language models available to your flows.
• Compliance - an EU AI Act risk assessment wizard that produces an auditable PDF and a tailored requirements checklist per workflow.

All four sit behind a single sign-on and are governed by role-based access control. See Functionality & Features for a deeper tour.

Compliance posture

Vextro is built to operate under EU regulatory regimes:

• GDPR - data subject rights, lawful basis, data minimisation, and privacy by design are baked into the platform.
• EU AI Act - every workflow can be classified using the in-platform wizard, which produces a risk level and a tailored requirements checklist.
• ISO 27001 / SOC 2 - the platform is structured to align with these frameworks. Achieving formal certification depends on your organisation's policies; the platform supplies the technical controls.

Detailed mapping is in Compliance Check.

High-level shape

Vextro is delivered as a self-contained platform that runs entirely inside your environment. The pieces that matter from a customer perspective:

• Web application - the user interface you log in to.
• Application API - the engine behind the UI; also the entry point for programmatic integrations.
• Workflow engine - executes workflow graphs and brokers calls to AI models.
• AI model engine - runs language model inference inside your environment.
• Knowledge index - stores the searchable representation of your knowledge-base content.
• Document store - keeps the original uploaded files and the PDFs the platform generates.
• Identity provider - handles single sign-on, role assignment, and session lifecycle.
• Secure gateway - terminates encrypted connections at the edge of the platform.
• Observability layer - captures execution traces, latencies, and any errors for audit.

A more detailed breakdown lives in Key Concepts.

Data flow

The full path of a single workflow run, from a user input to a stored result, is:

1. Request - the user submits an input from the UI, or an external system calls the platform API.
2. Authentication - the platform verifies the caller's identity through your single sign-on.
3. Authorisation - the platform checks role-based access and ownership (the workflow must belong to the calling user).
4. Execution - for synchronous runs, the workflow engine starts immediately; for asynchronous runs, the job is queued and a worker picks it up.
5. Retrieval (optional) - when a flow uses knowledge retrieval, the platform looks up the most relevant content from the relevant knowledge collection and binds it into the prompt context.
6. Inference - the workflow engine calls the configured AI model (local by default, an external provider if your administrator explicitly enabled one).
7. Trace - every model call is recorded for audit: prompt, model, latency, and any errors.
8. Persistence - the request, response, and status are stored.
9. Response - the platform returns the result; for asynchronous runs, the caller polls or subscribes to the job status endpoint.

Documents follow a parallel pipeline:

1. The user uploads a file via the Knowledge Base UI.
2. The platform stores the file and creates a knowledge source record in Pending state.
3. A background job picks the file up, extracts content, splits it into searchable chunks, and registers it in the knowledge index.
4. The knowledge source transitions to Processing and finally Completed.

AI models

Vextro is local-first by design. The default and recommended path is to run language models inside your environment, so prompts and documents never leave your perimeter.

• Local inference - language models such as Llama 3, Mistral, Qwen, and others run on your hardware.
• Embeddings for retrieval - also generated locally, configured per knowledge collection.
• External providers (optional) - major cloud AI providers can be enabled by your administrator, but only after explicit configuration. There is no implicit fallback to a hosted provider.

The Models page lists what is currently available in your environment. Workflow authors pick a model per node; there is no global default that silently applies.

Controls protecting your data

The platform enforces a layered control model. The essentials:

What to read next

• Key Concepts