Your next session already knows.
Zendavox keeps the context of a project as you work — what was decided, why, and what has already been tried — so the next session carries on instead of starting over.
The problem
You work on something with Claude, ChatGPT, Perplexity, or whichever AI assistant you use. You explain the project, the constraints, the decisions you already made and the approaches you already ruled out. You get somewhere.
Then the session ends. And the next one knows none of it.
So you explain it again. And again. Decisions get reopened because nothing recorded that they were settled. Work gets redone because nothing recorded that it was done. The longer a project runs, the more of each session is spent catching up rather than getting on.
What Zendavox does
Attach it to a project. As each session runs, Zendavox records the substance — not the transcript, but what matters: the decisions taken and the reasoning behind them, the facts that hold true, what was built or changed, and the documents that came with it.
The next session opens with that in hand.
Decisions, with the reasoning
Not just what was chosen — why, and what was rejected. A settled question stays settled.
Facts that stay true
The things you should never have to say twice: what it is called, where it lives, what it must never do.
What actually happened
What was built, fixed or changed, in order, so progress is visible across weeks rather than within an hour.
What a new session sees
This is the whole idea, and it is worth seeing literally. A fresh session on a project opens with something like this, before you type a word:
Example brief
Sessions so far: 14
Facts
deadline: filing due 14 March ·
constraint: no customer data leaves our servers ·
stack: Postgres, no ORM
Decisions already made
Q: Do we build our own billing?
A: No — use an existing provider. Rebuilt twice before and it
was never the differentiator.
Recent activity
Migrated the reporting queries · Fixed the
duplicate-invoice bug · Added the March filing documents
No re-explaining. No reopening the billing question for the third time.
Documents are part of it
A project is not only conversation. Things arrive — a contract, a records release, a batch of scans — and they belong to the same context as the decisions made around them.
Zendavox reads what arrives, works out what each item is, files it against the matter it belongs to, and makes people, organisations and addresses searchable across everything. Duplicates are caught before they are filed twice. An item referring to a record you do not hold is flagged.
And it counts. When you are owed documents by someone in no hurry to send them, the hard part is not reading what came — it is noticing what never did.
I have 7 of 43 — 36 still missing.
From a live public-records request. Zendavox took the total of 43 from the agency’s own document portal rather than anyone’s say-so. A folder can tell you it holds seven files. Only something that knows what was promised can tell you thirty-six are outstanding.
Start with the project you already have
You do not have to begin the day you install it. Point Zendavox at a project that already exists — sessions already run, files already accumulated, in sub-folders named inconsistently — and it works through the whole thing, filing what it recognises and setting aside what it does not.
Anything it cannot place, it holds and asks about rather than guessing. Answer once, and it remembers the words that set that document apart, so the next one like it is filed without being asked about. Your vocabulary becomes its vocabulary.
Who can see what
In a business, access is decided by document type rather than document by document. Management sets the rule once — procedures readable by all staff, personnel files by HR only — and whoever files an incoming document only answers what it is.
Managing access can be handed to an office manager or records clerk without handing over anything else. It never has to reach the person running the company, and every change is recorded.
Modules
Context is the foundation. On top of it sit modules you add to a basic subscription when you need them, each bringing the records and rules for a particular kind of work.
Customers
Who you do work for: people and businesses, how to reach them, and everything you have done for them.
Vendors
Who you buy from: your suppliers and the parts each one supplies.
Parts
The catalog: every part or item you use, with its number, description and cost.
Inventory
What you have on hand: stock levels for every part, what comes in and what goes out.
Invoicing
Getting paid: invoices for finished work, what has been paid, and what is still owed.
Work Orders
A job from start to finish: what was asked for, who it is for, the parts and labour that went into it, and the invoice at the end.
Board Records
Board meetings - who was there, what was decided and why, and what each meeting left someone to do.
Grants
Every grant you have applied for or hold, who funds it, the dates it brings with it, and what you have actually won.
Each module has its own page saying what it does and what it needs; a signed-in project adds one from your profile.
Three ways to use it
| Plan | Who it suits | What changes |
|---|---|---|
| Individual | One person, working alone | No job titles, no permissions, no access screens. Everything you put in is yours. |
| Small Business | A team where not everyone should see everything | Four job titles to start — Owner, Records Administrator, Supervisor, Staff — and one named person managing access. |
| Enterprise | Larger organisations | As many job titles as you need, and more than one administrator, because at that size a single one is a bottleneck. |
How it decides
You tell Zendavox a little about a project when you start it. From there, it learns as you go — what comes up as you write in a session, what gets decided, what gets built. The whole time, it is recording that context, without being asked and without waiting for the session to end.
Files work the same way. When you upload something to a session, Zendavox reads it and works out how it relates to the project. Where it is not confident, it stops and shows you the file and asks what it is. Answer once, and that type of document is recognised from then on — the next one like it is filed without being asked about again.
You can move between sessions as the work demands it — back to one from weeks ago, on to a new one, and back again — and every entry, in any session, updates the same project. So the picture you get back is never just the last conversation; it is everything that has happened so far.
The rules an assistant works from live in Zendavox’s own code and are served to it on every call, so every project runs on the same ones and an improvement reaches all of them at once, with nothing to repaste. The judging itself is done by the assistant you already use, following those rules. Zendavox runs no model of its own and sends nothing anywhere to be interpreted. Anything touching money, legal exposure or safety is held for a person to review regardless of how confident it is.
Questions people ask before they connect
These are the questions a careful person asks before pasting a key into their assistant. The answers are what the software does today, not what is planned.
Does Zendavox need my Claude, ChatGPT or Perplexity login?
No. It works the other way round. Zendavox gives you a key for one project; you paste that key into your assistant’s own connector settings; from then on the assistant uses it to identify itself to Zendavox. Zendavox never holds a login, password or token of yours, and has no way into your assistant account.
Does it see my chats or my files?
Your conversation stays where it is. Nothing reaches Zendavox except what your assistant explicitly records — a decision, a fact, a note that something was done, a preference — one short entry at a time, each as its own call. The transcript never leaves. A document reaches Zendavox only when you choose to file it there.
Where is my data stored, and in what form?
In a PostgreSQL database hosted by Supabase, as structured records: a decision is a row holding the question, the answer, the reasoning, who recorded it and when — not a copy of a conversation. Supabase encrypts customer data at rest and in transit, and every connection to Zendavox is over HTTPS: plain HTTP is redirected, never served.
What stops one project’s records reaching another?
The project you have open is part of every query Zendavox makes — named in the query itself, not filtered out of the results afterwards. A page has no way to return a row belonging to a different project, because it never asks for one. That is what protects you on the website and through your assistant’s key, which are the two ways anyone reaches Zendavox at all.
Underneath that, every table in the database carries its own row-level security policy, so anything arriving through Supabase’s interface rather than through Zendavox is held to one project by the database itself. Zendavox’s own server has to connect with an administrative role to run the site, and those policies do not apply to it — which is why the scoping in the queries, not the policy, is what we point at when you ask this question.
Can I run it on my own machine, licensed the way Office is?
Not today. Zendavox is a hosted service. The code runs against a local database for development, but it is not packaged as something you install and run yourself. If that is what you need, use the Enterprise form at the bottom of this page: write “run it on my own machine” in the box that asks what you are looking for, and press Request a conversation. It is a real option, and it is not built yet.
What does it add over my assistant’s own projects and memory? I can already tell a chat to read another chat.
Continuity rather than retrieval. Telling a chat to read another chat is a search you start: you have to know the earlier chat exists, point the assistant at it, and it hands back prose to be re-read every time. With Zendavox the session opens already knowing what was decided, why, and what was ruled out, with nobody pointing it anywhere. Three things follow:
- You never have to wonder whether it remembered. Retrieval fails quietly — forget to ask and the session carries on, confidently, on a wrong assumption. Continuity has nothing to forget.
- A decision is kept as a decision. With its reasoning and what it replaced — not a transcript that still contains what was later reversed.
- It is not tied to one assistant. The same record is read and written by Claude, ChatGPT, Perplexity and Claude Code. Change assistants and the project comes with you.
Is it about bigger file limits, or does it interpret my data differently?
Neither. Zendavox stores exactly what your assistant tells it and hands it back unchanged; all the thinking stays in your assistant. It is the difference between keeping notes and holding an institutional record.
Get started
Create an account, or log in.
Enterprise
Setting up Zendavox for a larger organization? Enterprise accounts are configured by our team, not signed up online. Tell us about your organization and we'll be in touch to get you set up.