Prepared data in
Accept structured rows inline or a secure organization-scoped dataset reference from another Zelvora capability or an external API client.
Back to products
Zelvora Reporting
Prepared data in. Complete report out.
Reporting is a reusable, data-agnostic rendering service. Other systems prepare the dataset; Reporting validates the request, renders the full result to Excel, PDF, or HTML, proves completeness, and stores the artifact securely.
- 3
- Output formats: XLSX, PDF, HTML
- 6
- Core backend components
- 1
- Rendering service reused portfolio-wide
- 0
- Silent partial outputs
Product vision
Operational reporting, not another BI stack.
Reporting is a multi-tenant renderer, template registry, background-job engine, completeness validator, and secure result service behind one stable API. Small datasets arrive inline; larger ones are referenced by organization-scoped dataset identifiers so tens of thousands of rows never pass through a model or a long-lived session.
The boundary is what keeps it reusable. Reporting does not fetch CRM data, run document extraction, decide which records match, or determine what a business should analyze. It turns a prepared result into a complete, secure, human-readable artifact - and nothing more.
Overview
Validate, render, verify, deliver.
Rendered asynchronously
Large reports run as jobs with progress state, retries, and resumable worker execution rather than long interactive sessions.
Completeness verified
Input and output row counts, failed rows, warnings, filters, omissions, and template version are reconciled before a report is called successful.
Delivered securely
Artifacts are stored under organization-scoped access control and returned as result metadata with short-lived access, never raw storage paths.
Output formats
One dataset, three artifacts.
The same prepared dataset renders to multiple formats while preserving a common request, template, and security contract.
Excel
The first high-volume renderer, using batch and streaming writes so large row counts do not exhaust worker memory.
Pagination, repeated headers, and branding handled through versioned templates with regression fixtures.
HTML
The same request, template, and completeness contract rendered for web delivery and downstream embedding.
Backend components
One rendering service for AI workflows and external clients.
Templates stay presentation-only. Versioning them makes output reproducible and keeps customer-specific layout logic out of renderer code.
Report API
Validate the request, tenant, format, template, and dataset contract before any work is queued.
Dataset reader
Load inline rows or secure references in controlled batches without owning the source system.
Job orchestrator
Queue long-running work, expose progress, and isolate generation from interactive sessions.
Template registry
Version layouts, supported formats, required columns, and customer branding - presentation only, no queries.
Renderers
Format-specific XLSX, PDF, and HTML implementations behind shared request and completeness contracts.
Completeness and result
Reconcile counts and warnings, persist metadata, and deliver secure, expiring result references.
Business use
Most useful when the data is already clean.
Because the renderer is uncoupled from any one domain, the same service serves finance, healthcare, operations, CRM analytics, and autonomous workflows.
Reconciliation reports
Render matched, unmatched, warnings, and summary sheets after another service performs the comparison.
Finance and audit packs
Turn prepared transactions, exceptions, or anomaly findings into controlled Excel and PDF deliverables.
Operational exports
Generate complete high-volume Excel reports from CRM, ERP, or connector datasets without sending all rows through an LLM.
Document-processing output
Convert structured Kursiv document-intelligence results into customer-facing PDF, HTML, or Excel outputs.
Autonomous workflow evidence
Curagentic or AI Employees generate a final human-review artifact after workflow execution.
External reporting API
Software products submit structured data and receive branded report artifacts through one reusable service.
Completeness and security
A silent partial file is worse than an explicit failure.
Completeness, tenant isolation, and scale controls are the product. Large files that exist but quietly omit rows are the failure mode this service is built to prevent.
A file is not successful merely because generation finished - completeness is a first-class quality contract.
Reporting returns complete=false when the requested data is not fully represented, with explicit failed rows, omissions, and warnings.
If a user asks for the full report, the system must not silently substitute a Top 10 or a sample.
Dataset references must belong to the caller organization and may expire; job and result identifiers are ownership-checked.
Generated artifacts are delivered through permission-aware, short-lived links rather than exposed storage paths.
Large workloads use worker queues with batch or streaming writes so generation scales independently of agent runtimes.
Benchmark framework
Test the renderer, not the upstream analytics.
The same fixed dataset and template run across benchmarks so measurements reflect rendering quality, completeness, latency, and resource cost - not differences in source queries or AI reasoning.
Completeness
Exact input and output row agreement, failed rows, omissions, and warning accuracy.
Generation latency
p50 and p95 end-to-end time by format and dataset size, with queue wait measured separately.
Scale
Largest stable row count and file size, worker memory use, and streaming behavior.
Format fidelity
Dates, currency, headers, pagination, repeated headers, clipping, and template regression checks.
Security
Zero cross-tenant access to datasets, jobs, results, templates, or generated files.
Unit economics
Worker time, storage, rendering, and result-delivery cost per report, plus template reuse share.
Exact row counts and large-dataset generation are benchmarked with fixed fixtures rather than manual spot-checking, and format-specific complexity - PDF pagination, Excel memory use, customer branding - gets regression fixtures of its own.
Market comparison
Win on composability, not on report authoring.
Microsoft, Jaspersoft, Telerik, and jsreport already provide mature rendering. Zelvora competes on source-system independence, AI-workflow integration, and provable completeness.
Power BI Paginated Reports
Established strength. Pixel-perfect paginated reporting, long multi-page reports, broad export formats, subscriptions, and deep Microsoft integration.
Reporting position. Stay narrower: a source-neutral rendering API for prepared datasets rather than a BI and report-authoring ecosystem.
JasperReports Server
Established strength. Mature report server with many data sources, rich formats, scheduling, and synchronous or asynchronous execution.
Reporting position. Differentiate through simpler dataset-reference contracts, AI-workflow integration, and explicit completeness metadata.
Telerik Reporting
Established strength. Embedded .NET reporting with a REST service, report viewers, and numerous output formats.
Reporting position. Target service-level reuse across products and runtimes rather than a .NET-centric reporting experience.
jsreport
Established strength. Developer-oriented open-source server with REST APIs and JavaScript or HTML templating across PDF, Excel, DOCX, and HTML.
Reporting position. The closest architectural competitor. Enterprise tenant controls and completeness verification are the meaningful difference.
Differentiation worth proving
- Prepared-data-in, artifact-out contract with no CRM, query, or OCR logic embedded inside Reporting.
- Explicit dataset references so large AI workflows never push tens of thousands of rows through a model session.
- Organization-scoped jobs, templates, results, and stored artifacts.
- Input and output row reconciliation with complete=true/false result semantics.
- One rendering service reusable by Brain, Kursiv, Curagentic, CRM analytics, and external applications.
Primary users
Finance, audit, operations, and healthcare teams, plus AI Employees producing human-review artifacts
Core workflow
Submit org context, format, title, template, and rows or a dataset reference; a job renders, verifies, stores, and returns an expiring result
Dataset model
Inline rows for small requests; secure references carrying tenant ownership, schema, row count, source artifact, and expiry for large ones
Out of scope
CRM fetching, business metric calculation, document extraction, and dashboard or BI functionality
Differentiator
Reuse: one engine for Brain, Kursiv, Curagentic, CRM analytics, and external clients, with completeness as a contract