Build on your own private AI, with the OpenAI API you already know.

from openai import OpenAI

client = OpenAI(
    base_url="https://dummydomain/v1",
    api_key="YOUR_API_KEY",
)

response = client.responses.create(
    model="Fast",
    input="Summarise this RBI circular in three points.",
)
print(response.output_text)

Change two things in code that already works with OpenAI: the base URL and the model name. Function calls, JSON schema output, images and streaming work as before.

Tiers are Instant, Fast, Smart and Genius. Your server tells you what each one can do.

QuickstartAPI reference

The API

OpenAI-compatible, plus tools that run on your server.

• POST/v1/responses

  Typed items, function calls that can return images, JSON schema output, typed stream events, the Agents SDK.

  Guide
• POST/v1/chat/completions

  Chat with images, tools and structured output for code written for chat completions.

  Guide
• POST/workspace

  Answers from your organisation’s documents, with sources and a found flag. Nothing stored.

  Guide
• GET/v1/models

  The tiers your key may use, with context windows and what each can do.

  Reference
• POST/chat

  The full assistant: attachments, audio, video, workspaces, web search and Office files.

  Tools
• POST/pdf, /ocr, /websearch

  Exact PDF text and tables, OCR for scans, and web search as single tools.

  Guide

For developers

Connect your app in five minutes

• Quickstart 5 min
• Ask your documents from your app Guide
• Responses API and Agents SDK Guide
• Build your own agents 9 examples
• Computer use: build a screen agent Guide
• Errors and limits Reference

For IT and administrators

Install, secure and run the server

• Hardware Checklist
• Installation Guide
• Users, keys and usage Guide
• Licensing and trial Guide

For teams at work

Get answers and documents from your own data

• Ask your documents Guide
• Make an Excel report Guide
• Minutes from a recording Guide
• Notes of a live meeting Guide

What you can build

Complete programs, each run against a ZenithAI server before it was published.

• Bills against the ledgerFunctions that return images

  The agent fetches each bill’s scan through your function, reads it, and checks it against what was typed.

• Ask your databaseFunction calling

  Managers ask in plain English; the model writes read-only SQL and explains the numbers.

• A multi-agent help deskAgents SDK

  A front desk hands each customer to an orders desk or a billing desk, each with its own tools.

• Bill and receipt photosImages and JSON

  Phone photos of bills become checked fields; doubtful ones go to a person.

• Meeting minutesAudio and Word

  A recording becomes decisions, owners and dates, and minutes as a Word file.

• Video checksVideo

  A site walk becomes a safety checklist, with the time each problem was seen.

What ZenithAI can do

Everything runs on your own server.

Site-locked chat

Answers only from the websites you trust, like rbi.org.in or gst.gov.in, with a banner that shows the lock.

How it works

Workspaces

Keep policies, manuals and reports in one place and ask questions about them in any chat.

Set one up

Audio and minutes

Attach a meeting recording, ask what was decided, and get minutes of meeting in Word.

Try it

Word documents

Letters, notices, memos, reports, offer letters and quotations on A4, in English or Hindi.

See the formats

Excel with live formulas

DD-MM-YYYY dates, lakh and crore grouping, charts, and every number checked before you download.

See the formats

XgenPlus Mail in chat

Connect your XgenPlus mailbox once, attach it to a chat, and ask about your mail, contacts and calendar. Read only.

Connect it

ZenithAI Agents

Ready-made tools in the web app, such as AI Inbox, Automation news alerts, Chess Coach, Mastery Coach, Note Taker and Designer.

See them