Platform Documentation

VisionAI is a cloud-based platform built on a microservices architecture designed for high availability and horizontal scaling. The system processes data through a series of configurable pipeline stages: ingestion, cleaning, transformation, analysis, and output. Each stage operates independently, meaning you can customize or bypass specific steps depending on your requirements.

The platform consists of four core modules that work together seamlessly. The Data Engine handles all connections, schema management, and transformation logic. The ML Engine runs model selection, training, evaluation, and deployment. The Automation Engine manages rule-based and AI-triggered workflows. The Presentation Layer powers dashboards, alerts, and report generation.

VisionAI uses OAuth 2.0 with Bearer tokens for API authentication. Each workspace generates unique API keys that can be scoped to specific permissions and expiration periods. For production deployments, we recommend using service account credentials with the minimum required permission set.

Generating API Keys

Navigate to your workspace Settings panel and select the "API Keys" tab. Click "Create New Key" and assign a descriptive name, permission scope, and optional expiration date. The full key is displayed only once during creation, so store it securely in your environment variables or secrets manager. If a key is compromised, revoke it immediately from the same settings panel.

curl -X GET https://api.visionai-platform.com/v2/datasets \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-Workspace-ID: ws_abc123"

SSO Configuration

Enterprise plans support Single Sign-On via SAML 2.0 and OpenID Connect (OIDC). To configure SSO, provide your identity provider metadata URL, entity ID, and certificate in the Authentication settings. VisionAI supports integration with Okta, Azure Active Directory, Google Workspace, and any SAML 2.0 compliant provider. Users authenticated through SSO inherit role mappings based on group attributes defined in your IdP configuration.

Rate Limits

API rate limits vary by plan tier. Starter plans allow 100 requests per minute, Professional plans support 500 requests per minute, and Enterprise plans have configurable limits up to 5,000 requests per minute. Rate limit headers are included in every API response:

X-RateLimit-Limit: 500
X-RateLimit-Remaining: 487
X-RateLimit-Reset: 1706832000

The Data Engine accepts structured, semi-structured, and unstructured data from over 120 sources. Ingestion can be configured as one-time imports, scheduled batch jobs, or continuous streaming feeds. The system automatically detects schemas, infers data types, and handles encoding conversion for files originating from different systems.

Supported Data Sources

Ingestion via API

Push data directly to VisionAI using our ingestion endpoint. The API accepts JSON payloads up to 10 MB per request. For larger volumes, use our bulk upload endpoint which supports multipart file transfers up to 50 GB. Both endpoints return a job ID for tracking ingestion progress.

POST /v2/datasets/{dataset_id}/records

{
  "records": [
    {
      "timestamp": "2026-01-15T14:30:00Z",
      "metric": "page_views",
      "value": 14523,
      "segment": "organic_search"
    }
  ],
  "options": {
    "dedup_key": "timestamp",
    "on_conflict": "update"
  }
}

Once data is ingested, the ML Engine processes it through several configurable stages. Automatic data profiling creates statistical summaries of each column, flagging missing values, outliers, and distribution characteristics. Feature engineering pipelines transform raw fields into model-ready features using encoding, normalization, binning, and interaction generation techniques.

AutoML Pipeline

The AutoML system evaluates multiple algorithm families (gradient boosting, random forests, linear models, neural networks) against your data and optimization target. It runs parallel training jobs with different hyperparameter configurations, using cross-validation to select the best performer. The entire process typically completes within 15 to 45 minutes depending on dataset size. Results include accuracy metrics, feature importance rankings, and a comparison table of all tested models.

Model Deployment & Monitoring

Deployed models run in isolated containers with auto-scaling based on prediction request volume. The monitoring dashboard tracks prediction latency, throughput, data drift, and accuracy degradation over time. When the system detects significant drift between training data and incoming production data, it triggers an automatic retraining workflow using the latest records. You can configure approval gates that require manual sign-off before a retrained model replaces the current production version.

The Automation Engine lets you create workflows that execute actions based on data conditions, schedule triggers, or model predictions. Workflows are built using a visual drag-and-drop editor that chains together input sources, condition nodes, action steps, and output destinations. Every workflow execution is logged with full audit trails, including input data, decision paths taken, and results produced.

Trigger Types

Available Actions

Workflows can execute over 40 built-in action types including sending email or Slack notifications, updating database records, calling external APIs, generating PDF reports, creating support tickets in Jira or Zendesk, pushing data to CRM systems, and triggering retraining jobs. Custom action scripts written in Python can be uploaded for specialized operations. Each action supports retry logic with configurable backoff intervals and failure notification routing.

The Presentation Layer provides a flexible dashboard builder with over 50 visualization widget types. Dashboards refresh in real-time or on configurable intervals, pulling live data from your connected sources and model outputs. Multiple dashboards can be created per workspace, each tailored to different audiences such as executive summaries, operational monitoring, or detailed analytical exploration.

Widget Types

Scheduled Reports

Configure automated report delivery in PDF, CSV, or Excel format. Reports can be sent via email to specified recipients on daily, weekly, or monthly schedules. Each report snapshots the current dashboard state, applies any saved filters, and includes a timestamp for audit purposes. Report templates support custom branding with your company logo, color scheme, and footer text. Distribution lists can be managed per report, with individual recipients able to unsubscribe from non-essential deliveries.

Official client libraries simplify API interaction by handling authentication, request formatting, pagination, retry logic, and type-safe response parsing. All SDKs are open-source, published to their respective package registries, and maintained by our engineering team with regular releases aligned to API version updates.

pip install visionai-sdk

from visionai import VisionClient

client = VisionClient(api_key="YOUR_API_KEY")
dataset = client.datasets.get("ds_revenue_2026")
predictions = client.models.predict(
    model_id="mdl_churn_v4",
    data=dataset.sample(1000)
)
print(predictions.summary())

npm install @visionai/sdk

const { VisionClient } = require('@visionai/sdk');

const client = new VisionClient({ apiKey: 'YOUR_API_KEY' });
const results = await client.datasets.query('ds_revenue_2026', {
  filters: { segment: 'enterprise' },
  limit: 500
});
const insights = await client.analyze(results);

// Maven dependency
<dependency>
  <groupId>com.visionai</groupId>
  <artifactId>visionai-sdk</artifactId>
  <version>1.5.3</version>
</dependency>

VisionClient client = VisionClient.builder()
    .apiKey("YOUR_API_KEY")
    .build();
Dataset dataset = client.datasets().get("ds_revenue_2026");

VisionAI emits events at key points in the data pipeline and model lifecycle. You can subscribe to these events by registering webhook endpoints in your workspace settings. Each event delivery includes a cryptographic signature (HMAC-SHA256) that your server should verify to confirm the payload originated from VisionAI.

Available Event Types

Webhook Payload Example

{
  "event": "model.training.completed",
  "timestamp": "2026-01-15T09:23:41Z",
  "workspace_id": "ws_abc123",
  "data": {
    "model_id": "mdl_churn_v4",
    "accuracy": 0.923,
    "f1_score": 0.891,
    "training_duration_seconds": 847,
    "dataset_rows": 245000,
    "status": "ready_for_deployment"
  },
  "signature": "sha256=a1b2c3d4e5f6..."
}

Common issues and their resolutions. If you encounter a problem not covered here, contact our support team through the contact page or use the in-app chat available in your workspace dashboard.