The private AI office.
On the record.
Minutes, research, decks, designs and agents for your whole team, running on one box in your office. Your files stay on the box, and every answer shows where it came from.
From ₹2,99,000 · indicative launch price, ex-GST · reservation refundable in full · the founding batch will be small







ANVE Office · render · unit in design
- Local modelQwen 3.8 27B on the box
- On the recordEvery call logged
- Real screensEvery product image is a screenshot of today’s build
- CheckedEvery capability on this page matched against our code
- 327automated tests in the platform
- ₹999reservation, refundable in full until you order
First for CA firms
The private AI office for CA firms.
Confidential files. A deadline every month.
Our first offices are CA firms of 15 to 50 people doing audit, GST, notices and ITR. Client files are processed on the box in the firm, and every AI call is logged to a named person. The whole suite comes with it: minutes, documents, decks, research and private assistants.
- Confidential files
Disclosing a client’s information without the client’s consent is professional misconduct.
CA Act 1949, Second Schedule, Part I, cl. (1) - A deadline every month
GSTR-3B is due on the 20th, every month. ITR dates in 2026: 31 July, 31 August, 21 November and 31 December.
GST portal · CBDT Circular 07/2026 - How firms cope today
So the file goes into a public chatbot, or it is done by hand. Financial Express reported in June that many CAs use public models such as ChatGPT, Gemini and Claude for audit, tax and compliance work.
Financial Express, 26 June 2026 (the reporter’s words, not ICAI’s)
“…remains within the institute’s ecosystem and does not leave the platform.”
On the record: every AI call has a name.
- LoggedEvery call through the gateway, by person, app and model.
- Sourced and ownedMeeting actions keep their source line and stay locked until a named person confirms.
- ApprovedTools and agents run only once the admin approves them.
Working today: the call log, the meeting trust gate and tool approval.
Each firm builds its own assistants.
In minutes, with no code: name it, write the instructions, tick the firm’s files it answers from, and keep it to yourself or share it with the whole firm.
New assistantExample
- Name
- Proposal writer
- Instructions
- Draft engagement proposals in our firm’s format, with our fee notes and scope language.
- Answers from
- ☑ Past proposals ☑ Fee schedule ☑ Firm profile
- Shared with
- The whole firm
Running today: minutes, deck, design, research-library and bill-entry agents, chaining up to seven approved tools, every call logged. Our CA agent library (GST notice reply, ITR preparation, TDS reconciliation and more) is built in a separate project and is not on ANVE yet; each agent joins only once it runs on the box.
About 1 lakh CA firms in India; 2,129 have six or more partners (ICAI data, October 2025). For a firm of up to 25 named people: ANVE Office, bought outright or rent-to-own.
Rent-to-own directly from ANVE, no third-party lender. ANVE Care: updates and support. Final price and the rent-to-own agreement are confirmed in writing before you order.
01 / Why now
Every office wants AI.
Few can send it their files.
Cloud AI asks you to send the work out. Offices bound by privilege, patient secrecy or official secrecy cannot. ANVE brings the AI in instead.
- Law firmsClient privilege under BSA s.132, and draft Supreme Court rules that ask lawyers to disclose AI use.See the rules →
- CA firmsThe CA Act bars disclosing client information without consent.See the rules →
- Hospitals and clinicsIMC Regulation 7.14 on patient secrets, and DPDP safeguards over vendors.See the rules →
- Government officesCCS (Conduct) Rule 11 on official documents, and CERT-In limits on third-party access.See the rules →
02 / The product
Intelligence your office owns.
One box on your network, for every named person in the office. Meetings, documents, research, decks, designs and agents in one place, with a record your compliance officer can read.
Watch the 50-second film →



Screenshots of today’s buildDemo data. Walk through every screen below.
03 / How it works
From a meeting to a finished screen.
Seven steps through the real software, captured from our build with demo data. Hover to pause, or pick a step.
Meetings
Hindi, English or both, transcribed on the box.
Record a meeting in the browser. Speech is transcribed on the machine itself, and every line keeps its timestamp, so any minute traces back to the moment it was said.
Screenshots of today’s build, with demo data.
Mixed Hindi–English speechTranscribed on this box04 / The suite
One box. Seven makers.
Everything editable.
Applications your staff already understand, behind one sign-in and one document library. Boards, decks and papers stay editable: change anything, then export to the formats your office already uses.
- Working todayMeetingsMeeting minutes
The meeting happened; now the decision stays. Hindi, English or both, transcribed on the box, every action with its quote and a confirmed owner.
- Edit before approval
- Quotes
- Owners

- Working todayDesign StudioUI/UX design
A Figma-style board. Screens from a brief, or from what a meeting agreed, with every layer editable.
- Edit every layer
- HTML
- React

- Working todayVellumPresentations
A local model writes every slide. Your team edits the deck, then exports it.
- Edit slides
- PPTX
- DOCX

- Working todayScholar-AtelierResearch
Papers drafted on the box and revised like a document: format, rewrite a selection, keep every version.
- Edit manuscripts
- Word

- Working todayStudioDocuments
Briefings, FAQs, study guides and timelines from any meeting or file in your workspace.
- From your knowledge
- Saved to the workspace

- Working todayChatPrivate GPTs
Assistants that answer from your approved files, with the source shown and the model named.
- Cited answers
- Logged calls
- Working todayStudioPodcasts
Two-voice audio briefings from approved material, with the narration voice clearly labelled.
- Script on the box
- Voice labelled
- Working todayMediaImages, vision and video
Generate images for slides and campaigns on the box’s own GPU, read scanned documents with OCR, ask about photos and diagrams, and search video feeds frame by frame.
- Image generation
- OCR
- Vision
- Video
- Working todayAPIFor your agents
An OpenAI-compatible endpoint on your network, so your own agents use the same private models.
- OpenClaw
- Hermes
- Coding agents
“Working today” means running in our current build. Screenshots use demo data.
05 / Agents
Agents that work across your office.
Coordinated. Accountable.
Ask once, in plain language. ANVE’s agent finds the right tools, runs every step across Minutes, Slides, Design and Research, and hands back the finished file, with every call on the record.
Agent runSample run · demo data · real tool names
“Make a six-slide deck from Tuesday’s programme review.”
search_toolsFinds the deck tools in the approved cataloguevellum · healthSlides is running on this boxvellum · generate_presentationSix slides from the approved minutesvellum · get_generation_statusDeck readyvellum · export_documentprogramme-review.pptx
5 tool calls · R. Sharma · Chat · local 27B · logged
Working todayAgents in the box
- Minutes agentDrafts decisions and actions with their quotes, checks every owner, and waits for a person to confirm.
- Deck agentTurns a meeting or a document into an editable deck in Slides, then exports it.
- Design agentLays out a screen in Design Studio from what a meeting agreed.
- Research agentWorks with the papers in your Research workspace, straight from Chat.
- Bills agentReads a GST bill and drafts the entry. Nothing posts until you confirm.
- Image agentGenerates images for slides and campaigns from Chat, on the box’s own GPU.
- Vision agentReads scanned documents with OCR and answers questions about photos, diagrams and video.
With your pilotAgents we build for your office
- Legal researchFinds precedent in your firm’s own opinions and judgments, every citation linked.
- Compliance checkReads a draft against your policies and the rules you answer to, and flags what to fix.
- Tender responseBuilds the compliance matrix for a tender and drafts the response from past bids.
- Action follow-upReminds owners before actions fall due and reports what slipped, straight from the minutes.
- Policy helpdeskAnswers staff questions from your own handbook and circulars, with the clause shown.
- Voice front deskAnswers callers and visitors in Hindi or English, built on AnveVoice.
Conversational AI
From the makers of AnveVoice.
AnveVoice, our voice AI for websites, starts its reply a median 487 milliseconds after a caller stops speaking, measured end to end. ANVE’s voice agents are built on the same craft: front desks, phone lines and helpdesks that hold a real conversation, in Hindi or English. ANVE itself was built by our founder, Adarsh Kant, with AI coding agents.
See AnveVoice →In the room
Made for the rooms you work in.
Chambers, boardrooms, studios and libraries. People use ANVE in the browser they already have, served by a box down the corridor.

PresentationsA deck drafted on the box, edited by your team, presented in the boardroom. 
Design StudioA Figma-style board for the screens your meeting agreed. 
MeetingsMinutes in Hindi, English or both, every owner confirmed. 
ChatAnswers from your own files, with the source and the model named. 
ResearchA paper drafted on the box, revised in a full editor.
Real screens · generated settings The rooms and devices are made with an image model; every screen is a real screenshot of today’s build with demo data.
06 / Integrations
Fits the office you already run.
ANVE works beside your existing suite. Files move in the formats everyone already uses, and live connections go through connectors your administrator approves and ANVE logs.
- Google Workspace
Decks and documents export as files that open in Google Slides and Docs.
- Slides
- Docs
- Zoho Workplace
The same files open in Zoho Show and Writer, and Zoho’s tools can connect through approved MCP connectors.
- Show
- Writer
- MCP
- Microsoft 365
PowerPoint, Word and PDF exports open in Office, exactly as your team expects.
- PowerPoint
- Word
- Any MCP server
Connect the tools your administrator approves. Each is fingerprinted, switched off if it changes, and every call is logged.
- Approved
- Fingerprinted
- Logged
- Your developers
Design Studio exports HTML and React, and an OpenAI-compatible endpoint serves your agents.
- HTML
- React
- OpenAI API
A connector to a cloud service sends that service only what it is asked to fetch or send, and every call is logged on the box. Product names belong to their owners; no affiliation is implied.
07 / The difference
On the record.
Personal AI computers promise an assistant that acts before you ask. An office needs the opposite: AI that never acts without a name on it.
- Every answer cited. Answers point to the page they came from. When the source is not there, ANVE says so.
- Every action owned. Meeting actions carry the quote that produced them, and nothing is approved until a person confirms the owner.
- Every call logged. Each model call and each agent tool call records who asked, from which app, and which model answered, on the box.
Programme review · Room 04 Example · demo data
- “Keep the first rollout to five systems until the attribution test is done.”Transcript
- Draft action: bring the evaluation set and acceptance threshold. Quote attached.Draft
- Owner confirmed by R. Sharma, programme lead.Confirmed
- Model call logged: R. Sharma · Minutes · local 27B model · on this box.Logged
08 / Hardware
Inside the box.
Pick a model, then scroll. The box opens, comes apart one part at a time, and closes again.
Specification at launch (may change before shipping) · renders · units in design
- Sealed, matte-black box for the office
- One cable to your network
- Nothing to install: everyone uses a browser
- Enclosure
- Cooling
- Graphics
- Processor
- Memory
- Storage
- Controls
- Models

- EnclosureSealed, matte-black box for the officeOne cable to your networkNothing to install: everyone uses a browser
CoolingLarge, slow fan over a copper heat-pipe coolerDesigned for a quiet office shelf
GraphicsNVIDIA RTX PRO 4000 Blackwell SFF Edition24 GB GDDR7 on the card770 AI TOPS (per NVIDIA; FP4 with sparsity)
Processor6-core AMD Ryzen 5 7600Runs the apps, search and transcriptionThe graphics card serves the models
Memory64 GB DDR5 system memoryPlus 24 GB GDDR7 on the graphics card
Storage1 TB NVMe SSDDocuments, models and the record of AI useFull-disk encryption
ControlsOffline switch cuts the network by handMicrophone switch with its own lightAmber status line, readable across the room- Models on the boxQwen 3.8 27B by default, on the 24 GB graphics cardA fast 4B modelWhisper large-v3-turbonomic-embed-text
09 / Engineering
What is in today’s build.
Checked line by line against our code on 7 October 2026.
- Speech
- Whisper large-v3-turbo, with WhisperX where installed, run on the box and chosen on our own Hindi measurements.
- Language model
- Qwen 3.8 27B by default, served on the box. Administrators can add others; answers from cloud models are labelled.
- Images and vision
- Image generation runs on the box’s own GPU through ComfyUI. Vision models read scans (OCR), photos and video frames.
- Agent loop
- Up to seven tool steps per request. Results enter as untrusted data, and a claim without a successful tool call is challenged.
- Minutes trust gate
- Each action needs a quote from the transcript. Approval stays locked until a person confirms every owner.
- Model gateway
- Every call is logged with the person, the app, the model, the tokens, and whether it ran locally.
- Tool host (MCP)
- Tools need an administrator’s approval. A tool whose definition changes is switched off. Every call is logged.
- One sign-in
- Vellum, Scholar-Atelier and Design Studio run as services on the box, reachable only through ANVE’s login.
- Answers that finish
- A reply keeps running if the browser disconnects, and is waiting when you come back.
- Tests
- 327 automated tests in the platform repository.
- Telemetry
- None. ANVE’s own software sends no usage data to ANVEAI.
10 / Lineup
One for you. One for your office.
One professional, at home or in chambers
ANVE One
₹2,99,000 ex-GST
- Users
- 1–3 named users
- Hardware
- AMD Strix Halo class
- Memory
- 64 GB unified memory
- Models
- 27B-class (Qwen 3.8 27B) plus a fast 4B model
Firms, clinics and departments
ANVE Office
₹4,49,000 ex-GST
or rent-to-own, ₹13,999 a month × 48 months ex-GST · care included · indicative
- Users
- up to 25 named users
- Hardware
- NVIDIA RTX PRO 4000 Blackwell SFF Edition (24 GB GDDR7) · 6-core AMD Ryzen 5 7600
- Memory
- 64 GB DDR5 system memory · 1 TB NVMe SSD
- Models
- 27B-class models (Qwen 3.8 27B by default) on the 24 GB graphics card
Larger offices, labs and PSUs
ANVE Office Pro
₹9,49,000 ex-GST
- Users
- up to 100 named users
- Hardware
- NVIDIA RTX PRO 4000 Blackwell (24 GB)
- Memory
- 128 GB system memory
- Models
- 27B to 35B-class models, served to many users at once
Departments and campuses
ANVE Estate
On quote
- Users
- 100+ named users
- Hardware
- NVIDIA RTX PRO 5000 or 6000, configured to order
- Memory
- Configured to order
- Models
- Larger models on request
Indicative launch prices, excluding GST. Final prices are confirmed after our own engineering units are validated; pre-booking keeps its 15% benefit until then. Concurrent-user capacity is published after benchmarks on the shipping unit. Compare with Ghost, DGX Spark, Mac Studio and others →
11 / Pre-order
Reserve ANVE for ₹999 / US$12.
Pay only the reservation today, through secure Razorpay checkout. You get 15% off the final announced price, lifetime software updates and a founding-member badge, and a full refund any time before you confirm an order in writing.
Reserve now.
Decide after the final price is revealed.
Your reservation holds the 15% benefit, lifetime updates and your founding-member badge. It does not commit you to buy the final product, and it remains fully refundable before written order confirmation.
Reserving holds your place; you choose buy or rent-to-own when you order. The 15% pre-booking benefit applies to the final announced price as the pre-booking terms state, and how it applies to rent-to-own is confirmed in writing before you order.
Want to see it first? Book a free 20-minute demo call, or read the pre-booking terms.
Not ready to reserve? Get the build notes and launch news by email: what works, what slipped, and the final price when it is announced.
What leaves the box
Nothing, unless you choose it.
- Meetings, documents, chats, agents and the record of AI use stay on the box.
- ANVE’s own software sends no usage data to us.
- Visible exceptions: podcast narration uses a cloud voice; cloud models and connectors only if an administrator adds them, labelled and logged; Research looks sources up online when asked.
If we disappear
Your box keeps working.
- A perpetual licence for the software delivered with every unit.
- No remote kill switch.
- Written into every order, not just said in an interview.
Questions offices ask.
- Is ANVE made for CA firms?
- CA firms of 15 to 50 people are our first offices. Client files are processed on the box in the firm, every AI call is logged to a named person, and the firm builds its own private assistants from its own files. The full suite comes with it. Our CA agent library is built in a separate project and is not on ANVE yet.
- Does ANVE need the internet?
- No. Meetings, documents, chats, agents and the record of AI use stay on the box. A few things can reach out, each visible and optional: podcast narration (a cloud voice), cloud models or connectors an administrator adds (labelled and logged), Research’s online source lookup, and downloads during setup.
- What can ANVE’s agents do?
- From Chat, ANVE’s agent finds the right tools and runs every step: it makes decks in Slides, generates images, reads scanned documents with OCR, lays out screens in Design Studio, works with your Research library and drafts bill entries, with every tool call logged. During your pilot we build the agents your office needs, such as legal research, tender response or a voice front desk.
- Can we edit what ANVE makes?
- Yes. Design Studio is a Figma-style board where every layer can be changed and exported as HTML or React. Decks open in Vellum’s editor and export to PowerPoint, PDF or Word. Research papers are revised in Scholar-Atelier’s editor and export to Word or PDF.
- Does it work with Google Workspace or Zoho?
- Yes. Decks and documents export to formats that open in Google Slides and Docs, Zoho Show and Writer, and Microsoft Office. Live tools connect through MCP connectors your administrator approves, and every call is logged on the box.
- Who can see what?
- Every person signs in with their own account. Meetings and private documents are visible to their owner, shared documents to the office, and administrators decide which apps each person can use.
- Can we show what the AI did?
- Yes. Every model call and every agent tool call is logged with the person, the app and the model, and every confirmed meeting action records who confirmed it.
- What happens if ANVE shuts down?
- Your box keeps working. Every order includes a perpetual licence for the software delivered with it, and there is no remote kill switch.
- Does it help with the DPDP Act?
- It removes the biggest risk: personal data is not handed to an outside AI processor. You remain the data fiduciary; we publish exactly what leaves the box so your compliance team can sign off.
- What does a pilot include, and what does it cost?
- A 20-minute demo call is free. The pilot is paid: we install ANVE in your office for your team and your own files, build the first agents with you, and agree the fee in writing before anything is installed.
- When can we buy one?
- Pre-orders are open: reserve for ₹999 / US$12 on this page, refundable in full. Our own sealed unit is in design, pilots come first, and final prices are confirmed after engineering units are validated. Indicative prices start at ₹2,99,000, and ANVE Office can also be rent-to-own at ₹13,999 a month for 48 months, ex-GST, with care included (indicative).
Keep your office’s intelligence in your office.
The first boxes are in design, and the founding batch will be small. Pre-orders are refundable in full.
