Built for companies of 10 to 150 people

AI that works inside your own files

Ofivia keeps your company documentation as Markdown, in a folder that belongs to you. On top of it runs an agent that reads it, writes to it, and tells you which note each fact came from.

  • 10,000 credits on the house
  • No credit card
  • Every feature from day one

A department agent handling three real requests inside a company’s documentation: what was agreed with a supplier, a process manual drafted from meeting minutes, and one area’s open tasks. Every answer carries the note it came from.

Built for

Companies that live off what they have written down, and today have it scattered across Drive, WhatsApp and the memory of two people.

  • clinics and health networks
  • law firms
  • consultancies and accounting
  • importers and distributors
  • agencies

Ofivia answers with what your company already wrote down

It does not invent a plausible answer. It searches your company vault, opens the notes that matter and shows you which ones they were.

in the board minutes. in the process manual. in the proposal you sent out in March. in the contract you signed two years ago. in the note somebody left and nobody opened again

Why it feels different

Four decisions you cannot walk back later

Almost everything an AI tool promises can be shipped in a later release. These four are settled in the architecture or they are never settled at all.

textiles-andinas/

Imports/

suppliers/

textil-kunshan.md

customs/

Production/

manuals/

goods-receiving.md

It works inside your files

Every company gets its own folder on disk, mode 0700, isolated by the filesystem. Inside it there is Markdown and the originals you uploaded. The index and the graph rebuild from that folder, never the reverse.

How the vault works

Upload 4,312 documents

0 credits

Index and reindex

0 credits

Search the whole archive

0 credits

One working turn with the agent

≈ 190 credits

Indexing and search cost nothing

Indexing runs on your server with a local model, and search is a query against your own database. Upload twenty years of documentation and search all day: the meter does not move.

Search and graph

Marcela

Head of Imports

runs her branch and everything under it

Andrés

Sales

sees Sales, nothing from Executive

Outside accountant

Read only

cannot write, and neither can her agent

Permissions are the ones your people already have

A role in the company plus a seat in each department. An area lead administers their branch and everything under it. A department agent works inside that subtree and never sees the rest.

Departments and permissions

$ rsync -a ofivia/vault/ ~/backup/

5,575 .md files copied

112 MB, not one proprietary format

opens in any text editor

You can take the folder and leave

It is Markdown files and a standard Postgres database. Drop Ofivia tomorrow, copy the folder, lose nothing. And if you would rather not depend on us at all, one command installs the whole thing on your own server.

Portability and self-hosting

Comparison

Four jobs, four ways of getting them done

Pick the job and see who actually does it. The other three are good at what they are good at, which is why the whole row is worth reading.

“What did we promise this client back in 2024?”

Ofivia

Searches by meaning and by exact term at once, across every document the company has, and hands back the note and the paragraph. Nobody has to remember what the file was called.

Yes

ChatGPT

Answers beautifully about whatever you paste or upload into the conversation. The PDF nobody has opened in two years is not in there, and nobody thinks to put it there.

Halfway

Notion AI

Finds whatever lives inside Notion. What stayed in Drive, in an email thread or in a scanned contract is out of reach.

Halfway

A shared folder

It is all in there. Finding it depends on somebody remembering the filename, or having the afternoon free.

No

The long versions live at Ofivia vs ChatGPT and Ofivia vs Notion AI .

How it works

Three steps, and only the third one spends anything

  1. Upload your documentation

    Drag in whatever you have: PDFs, Word files, spreadsheets, minutes, manuals, contracts. It becomes Markdown, the regenerable junk is stripped, and it gets indexed.

    Uploading and indexing consume no credit. Indexing runs on your own server.

  2. Lay out your departments and permissions

    Draw the real org chart: Executive, Operations, Sales, whatever you have, as deep as it needs to go. Each person arrives with a role and a seat.

    Every department is born with its folders and its main note.

  3. Give the agent work

    Ask a question, hand off a draft, ask what is still open. The agent searches, reads, writes inside the department and shows you every tool it runs.

    This is where credit starts counting, and you see exactly where it went.

You are in charge

Everything the agent does leaves a trace

An agent that writes into company documentation is only useful if you can audit it afterwards. Here is what gets recorded, exactly as it is built today.

  • Every call is metered

    The meter rolls up by day, by model and by process, at what the provider charged rather than at an estimate. Billing is idempotent: a retry never counts twice.

  • Every tool call is audited

    Every run the agent makes writes a hash-chained row. Edit one or delete one and the chain stops matching, which is the point. What is stored is the hash of the arguments, never the arguments.

  • Secrets are never displayed

    Your company keys are stored encrypted and the interface lists them by name. Nobody sees a value on screen, the company owner included.

  • A person approves the credit

    A company can request more credit and watch its budget, but it cannot approve itself. The request goes to a person who grants or denies it, and that decision is final.

How the security is built

The meter reports, it does not cut anything off. If a month runs over budget you see it with a name, a date and a process attached, not on next month’s invoice.

Second opinion

Ask another AI about Ofivia

We wrote the question for you, weaknesses included. It opens in a new tab and the answer never passes through us.

What is Ofivia (ofivia.com)? Tell me how it differs from ChatGPT and Notion AI, where it is strong and where it is weak, and what kind of company it makes sense for. Be critical.

Questions

What people ask before they decide

See all questions
What happens to my documents if I stop using Ofivia?
You take them with you. Everything lives as standard Markdown files inside your company folder, next to the originals you uploaded. Copy the folder and open it in any text editor. There is no proprietary format to export out of.
Does Ofivia train models on my company documentation?
No. Your documents are indexed so they can be retrieved, and that indexing runs on the server where the system is installed. The data processing agreement puts the obligation in writing.
What does it cost to upload and index the entire archive?
Nothing. The model that builds the index is local, so uploading, indexing, re-indexing, searching and walking the link graph consume no credit. Credit is consumed by the agent when it works.
What exactly is a credit?
One credit is a tenth of a US cent of real spend with the model provider. A working turn with the agent averages about 190 credits measured on this installation, so 25,000 credits a month is roughly 130 working conversations.
Can I open the account myself?
No, and that is deliberate. You leave your company details and a work email, confirm the email, and a person creates the company with its departments and its starting credit before writing back. It takes under one business day, and it keeps an account from being born misconfigured.
Can I run it on my own server?
Yes. The full install is one command and everything sits under a single directory: database, vault, graph and backups. Daily backups use hard links, so each day costs on disk only what actually changed.

Start free. Pay when it earns it.

One credit is a tenth of a US cent of real spend with the model provider. No per-seat licenses and no minimums.

Trial

10,000 credits on the house

No card, no expiry. Enough to try it against your own real documentation before deciding anything.

Basic, US$80 a month

40,000 credits a month

About 210 working conversations, shared across the whole company. If a month runs short, you top it up.

Pro, US$150 a month

90,000 credits a month

For the whole organization, with scheduled automations, encrypted secrets and white label.

Per-company vault at mode 0700 Markdown you can walk away with Semantic and exact search at once Link graph across your notes Indexing consumes no credit One agent per department Tool calls visible as it works Memory that is dated and sourced Roles that inherit downward Hash-chained tool audit Encrypted secrets, listed by name Scheduled automations Voice notes, transcribed White label per company One-command install Daily backups with hard links

Bring your documentation and see what it answers

We hand you starter credit so you can try it on real minutes, manuals and proposals. No card and no sales call: you leave your details, a person sets the company up and writes back.

  • Starter credit on the house
  • No credit card
  • Every feature included