Platform

An AI operating layer for your organisation.

ZenithAI combines a private assistant, a governed knowledge layer, department workspaces and document generation into one platform — installed inside your security boundary, working across the material your organisation already has.

Sec. 01 The assistant

A private assistant that works like your best analyst.

Every user gets a capable conversational assistant for questions, analysis, drafting and coding, running on models hosted in your environment, with full chat history you can search (Ctrl+K finds any word in your questions and answers), session pinning, More styles to rewrite a drafted email or reply in other styles, light and dark themes, and transparent response metadata: which tier answered, how fast, at what length. Every turn runs through the ZenithAI Intelligence Framework — a deterministic reasoning pipeline, not a single model call.

  • 01Automatic tier routing — complex analytical prompts escalate to deeper reasoning tiers on their own
  • 02Deterministic calculation checking — numeric claims are verified before they reach the user
  • 03Multilingual by nature — Hindi and other Indian languages via natively multilingual models; mathematical notation rendered properly
  • 04Multimodal input — images, audio and speech, with local transcription

Sec. 02 Document intelligence

From files to defensible answers.

Upload policies, contracts, registers and reports directly into a conversation. ZenithAI chooses the right strategy for each file — and shows its evidence.

Prose documents

Cited retrieval

Long documents are indexed passage-by-passage using locally hosted embeddings. Answers cite the pages they came from, so any statement can be verified at the source before it is acted on.

Spreadsheets & tables

Verified computation

Questions about tabular data are answered by generated analysis code executed in a hardened sandbox — restricted language surface, resource limits, no network — and badged “✓ Computed”. The figure is calculated from the register, not guessed by a model.

Scanned material

Local OCR

Scanned PDFs and images are read by a locally hosted vision OCR model. Legacy archives and signed correspondence become searchable without any external OCR service.

In every chat

Attachments you can open again

A file attached in chat opens again from its card: images, PDFs, audio, video and text in a viewer, Word, Excel and PowerPoint as their first five pages, with a Download button for the original. When the administrator switches keeping on, files stay on your own server for 90 days by default or the number of days the administrator sets, and only the person who attached a file can open it. Chat reads up to about 48,000 characters of each file; whole collections belong in a workspace. Kept attachments and the viewer come with the next installer, after 4.6.0. Attachments in chat.

For pipelines

Deterministic PDF extraction

A separate LLM-free API extracts text and tables from PDFs with page structure intact — for workflows that need structure without a model in the loop.

Sec. 03 Workspaces & knowledge

Institutional memory, scoped and governed.

Each workspace carries its own persona, pinned facts and knowledge base — a procurement desk, a policy cell, a legal review room. Workspaces are private to their creator by default; administrators deliberately publish shared, organisation-wide workspaces.

Isolation is enforced at every endpoint: another user’s workspace doesn’t just refuse access — it doesn’t acknowledge existence. Follow a workspace from first document to answer →

  • 01Durable facts — tell a workspace to remember a threshold, a convention, a standing instruction; every future conversation starts knowing it
  • 02Per-workspace knowledge bases — reference documents injected into any conversation on demand
  • 03Knowledge from public websites — import a public URL and it becomes part of the knowledge your teams talk to, extracted and indexed on your hardware like every uploaded document. Website Knowledge →
  • 04Deliberate publishing — private by default, shared only by an administrator’s explicit decision
  • 05An asset that compounds — organisational memory grows with use and stays inside your walls
  • 06Answerable from your own software: the same workspace answers, with sources and a found flag, through one API call from your intranet, help desk or apps, with the asker’s own access. Enterprise Knowledge API →

Sec. 04 Document generation

Deliverables, not just replies.

Ask for a letter, a report, a board deck or an analysis workbook, and ZenithAI produces the actual file — formatted by a built-in layout and quality engine, ready to download from a card in the conversation.

Offer letters, memos, GST invoices, payroll and EMI schedules, with example prompts in English, Hindi and Hinglish: Word and Excel documents →

  • 01Word — letters, office notes, proposals, BRDs and structured reports
  • 02PowerPoint — presentation decks with charts and consistent visual design
  • 03Excel — analysis workbooks driven by real computation, and dashboards from your own spreadsheet with KPI cards, charts and a data quality sheet
  • 04Governed like everything else — bounded retention and size limits

Sec. 05 Connected capabilities

Optional modules that reach outward — on your terms.

Some work needs the outside world. These modules are individually enableable, logged, and designed so consequential actions always require a human decision.

ZM-01Web researchCurrent-affairs and research questions answered through a search stack hosted on your own infrastructure — not a third-party search API — with sources cited.Optional module
ZM-02AI InboxAn agent that connects organisational mailboxes to read, sort and summarise mail and draft replies. Nothing is sent without a person pressing Send. Live since 15-10-2026; not in installer 4.6.0. AI Inbox docs.Optional module
ZM-03AI-assisted callingOutbound calls proposed by the assistant — follow-ups, reminders, verifications — dialled only on explicit, contemporaneous human confirmation. Requires organisational calling infrastructure, provisioned during scoping.Optional module
ZM-04Domain intelligenceOn-demand intelligence reports on external domains and organisations — registration, infrastructure, web presence, reputation — assembled by your own deployment.Optional module

Sec. 06 For your engineers

A platform, so also an API.

Everything user-facing is available programmatically: chat, search, PDF extraction, OCR jobs and domain intelligence over a REST API secured with managed keys and usage tracking. Build ZenithAI into portals, service desks and internal systems your teams already use. The /v1 API is OpenAI-compatible, including the Responses API and function calling on the Fast, Smart and Genius tiers, and POST /workspace answers from your documents. Computer use agents read screenshots on the Genius tier and propose clicks, typing and scrolling that your own program performs (live on our servers since 25-10-2026, not in installer 4.6.0).

  • 01Database querying — natural-language questions over enterprise databases, schema-aware and read-bounded, scoped per workspace
  • 02Single sign-on embedding — ZenithAI embedded into existing portals with signed, short-lived tokens
  • 03OCR document viewer — full-fidelity view of scanned documents with searchable output
  • 04Administrative console — user and system oversight with audited access

Sec. 07 For administrators

Run it like your own system.

Administrators see who uses ZenithAI, how, and how much space it takes, from the same browser app. Every account change and every view of someone’s chats or files is written to the audit log.

API & Usage, Live Snapshot and Storage are live on ZenithAI and come with the next installer; they are not in installer 4.6.0. Users and administration docs →

  • 01Users and roles: add people one by one or from a CSV file, User Managers, suspend and archive with a recorded reason
  • 02Internet per person: switch web access off for anyone; nothing goes to the internet for them, even on request
  • 03API & Usage: every API key with its owner and last use, a switch that lets a key add documents, the Top 10 users and why requests failed
  • 04Live Snapshot: the last 10 minutes of requests, tokens and active users, by chat, API, workspace and Agents
  • 05Storage: disk used, chat attachments, files expiring soon and the largest users, with checked actions such as changing how long attachments are kept

Questions

The platform, answered.

What is ZenithAI?

An enterprise AI platform by Data Ingenious that runs inside your own infrastructure. It gives every user a private assistant, answers from your documents with citations, handles spreadsheet questions by running real calculations, and organises knowledge into governed workspaces.

Is it a SaaS product?

No. It is deployed on your on-premises GPU servers, in your private cloud or in an isolated network. The models run on your hardware.

How does it answer questions about our documents?

Long documents are indexed passage by passage with locally hosted embeddings, and answers cite the pages they came from. Questions about spreadsheets are answered by analysis code run in a sandbox, so the figure is calculated from the register, not guessed.

Can it read scanned files?

Yes. Scanned PDFs and images are read by a locally hosted OCR model, so old archives become searchable without an external OCR service.

Who can see a workspace?

Only its creator, until an administrator deliberately publishes it. Another user's request for a private workspace is refused without confirming that it exists.

Does it work in Hindi?

Hindi and other Indian languages work through natively multilingual open models.

The next step

See the platform on your material.

A demonstration takes under an hour and uses realistic documents for your sector.