What's new

New features added to Verbatim AI, most recent first. Only new capabilities are listed
here — smaller improvements, fixes and behind-the-scenes work are left out.

Last release : v0.2.1

---

October 2026

🎟️ Access tokens, scoped and short-lived

Mint an API token for a front-end or a script without publishing an RSA key. The key
page now has an Access tokens section: pick the scopes the token may use — corpus:read,
doc:create, post:read… — and how long it lives, from an hour to a year. It is shown
once, with a button to copy it; send it as the X-Access-Token header of your API calls.

Every token of your organization is listed with what it carries and when it expires, and
Revoke in its menu cuts it off at once. See API keys.

---

September 2026

🧠 Pick a model by what it does, not by its name

The two model fields of an agent now open a list of the models Verbatim AI serves, each
with its provider's logo, its display name and a line on what it is good for —
Platform default at the top, as the choice that keeps following ours. Before, both
fields were a menu of bare identifiers, which said nothing about what picking one would
change. An agent's page names its models the same way, so what you picked is what you read
back, and every agent card carries the model it answers with — logo and name — whenever it
pins one.

🔖 The Stories you keep coming back to

Bookmark a Story and find it again in one press. The bookmark in a Story's header marks
it as a favourite; its row on the Story screen then carries the mark beside its name,
and the bookmark above the list keeps only the Stories you marked.

It narrows alongside the Corpus filter rather than instead of it, so "my favourites in
this Corpus" is one list. Pressing the header bookmark again takes the mark off — a
favourite is a shortcut to a Story, not a state the Story is in.

🗂️ The Corpus space loads as you scroll

Your Corpus list no longer stops at a fixed batch. It loads twenty-five at a time and
asks for the next ones as you reach the end, so an organization with a long shelf of Corpus
sees all of them, and opens as quickly as one with two. Pull the list down to refresh
it after a colleague has created a Corpus.

📚 A history you can actually search

The Story screen sorts, filters and pages. A count of everything that matched sits above
the list, with an arrow that flips between most recent and oldest first, and a funnel
that narrows the history to a single Corpus. Each row now names the Corpus its Story
reads from, one chip each, and a click opens it.

A Corpus page shows its five most recent Stories; See more under them opens the whole
list, already narrowed to that Corpus and carrying the same controls.

The list also loads as you scroll now, twenty-five Stories at a time, instead of
fetching a fixed batch up front — so a long history opens as quickly as a short one.

🔎 Find everything like this document, in one tap

Your file list is now navigable by what its documents are, not just by their names. Under
each name sit the chips the document is filed under — its format first, then its tags.
Tap one and the list narrows to everything else like it: every spreadsheet, every document
tagged legal. The funnel button shows the criterion in force, and Clear brings the whole
Corpus back.

The same chips sit beside the document id in its Details, and do the same thing from there.

Each row also shows the first page of the document itself rather than a symbol shared with
every other file of the same format — far quicker to find a document by, in a list of twenty.
Documents with no page rendered yet keep their format symbol.

🗂️ Four more formats

Spreadsheets and plain text join the Corpus. Alongside PDF, Word and PowerPoint, you can now
upload Excel .xlsx, Markdown .md, plain text .txt and .csv files, and the
filter offers all seven formats.

🧩 See what your Corpus actually reads

A document now tells you what it is made of. Details, in a file's ⋮ menu — or the
ℹ️ button in its preview header — opens its full record — when it arrived, what it weighs, how many pages and passages it holds, and the
storage it really occupies once the page images, the text conversion, the summary and the
embeddings are counted alongside the file itself. Its first page sits beside all that, so you
can tell at a glance which file you opened.

The record also says what reading the document cost — the tokens spent writing its summary —
and the chunking settings it was cut with, so you can check what a file was actually ingested
with rather than what you meant to send.

From there, View chunks opens the passages the document was cut into. Your Corpus does not
answer from whole documents, it answers from those passages — so this is the page to open when
an answer quotes a file and gets it subtly wrong. Each one shows the pages it covers, its text
as stored, and its metadata.

You can correct a passage in place — a mangled character, a name to remove — or delete
one to take it out of the index without touching the document it came from.

⚠️

Correcting a passage does not re-index it: it keeps being found on the wording it had.

For a real rewrite, replace the file instead.

📬 Email your Corpus

A Corpus can now collect mail. Open a mailbox on it and the platform mints an email
address; everything sent there is ingested, with no upload step at all. Forward a thread, put
the address in copy of a supplier, or hand it to a scanner — what arrives becomes knowledge
your assistant can answer from.

Each mailbox chooses what it keeps out of every message: the body and the attachments,
the attachments only — the files people send, not their covering note — or the body
only. A switch on its card starts and pauses the collection without giving the address up,
and a new mailbox always starts paused so you can check it before the first message lands.

Open a mailbox to browse an inbox of what it collected: who wrote, what about, the text
of the message and the files it carried. Every message says whether it reached the Corpus,
and an attachment that did opens its preview in one tap. One that did not is marked
Not kept — which is how you find out why a file somebody sent is missing.

Mailboxes live in the Corpus page, under their own section. See
Mailboxes.

---

August 2026

📊 Your consumption, right on the dashboard

The Statistics card has grown into a panel. The two numbers it carried — the prompts
you have run and the tokens they cost — now head a panel that charts the last 30 days
behind them: tokens consumed, day after day, on the home screen itself. A quiet spell or a
jump in consumption is visible without opening anything.

On a wide window the panel goes further and charts activity — the stories and prompts
opened each day — and storage used. On a phone it keeps to the tokens, and Show more
opens the full report either way. See
The dashboard.

💡 Contextual help, on your terms

The help panel is now yours to open and close. Every screen carries a help button in
its header: tap it to slide the panel in, tap it again — or the ✕ in the panel's own header —
to send it away. On a wide window the panel settles beside the page; on a narrow one it
slides over it like a menu drawer, and a tap on the page behind puts it back.

And it remembers. The state you leave it in follows you from screen to screen, and it is
still there the next time you open Verbatim AI. Until you touch the button, the panel opens
by itself where the window is wide enough to hold it, and stays out of the way where it is
not.

🤖 Put an agent to work from the agent's own page

Choosing who answers you now takes a button that says so. Open an agent from the Agents
list and use Use this agent, pinned to the bottom of its page. The agent already
answering shows No more use there instead, which hands your questions back to the one
badged Default.

The Agents list briefly did this itself, on a tap on a card. It was quick, but it put
changing the agent behind every later question one stray tap away — so tapping a card is
back to simply opening it, and the choice moved to a place you cannot reach by accident.
The list still shows which agent is at work: it carries an Active chip, alongside
the Default badge when the two are the same agent.

The 🤖 button in a Story's header and the Agent section of your profile are unchanged.
See Agents.

🕘 See what a Story reads from, at a glance

Every Story now wears the same icon, badged with the number of Corpus it searches. A
Story over three Corpus says so in the list, in its header and everywhere else it appears —
where before it borrowed the icon of a single Corpus and said nothing about the others.

Open a Story and a row of chips at the top names each Corpus it reads from; click one to
go straight there. See Stories.

🔗 Ask one Story across several Corpus

A Story is no longer tied to a single knowledge base. The new ➕ button on the Story
screen asks which Corpus to read from, and you can tick as many as you like. Ask your
contracts and your invoices the same question and get one answer drawn from both — still
naming the documents it used, whichever Corpus they came from.

The Start a story button on a Corpus page is unchanged: it opens straight onto that
Corpus, as it always has. See Stories.

🍴 Start from a platform agent

Fork any of Verbatim AI's own agents into one of yours. Platform agents are read-only,
which used to make their configuration something you could look at and never build on. Open
one and press Fork: the create form arrives carrying every one of its settings, with
only the name left empty for you to fill in.

Change as much or as little as you like before saving. Your copy joins your agents fully
editable, even though the agent it came from never was, and the original is left exactly as
it was. See Agents.

🤖 Pick the agent your questions run under

Choose which agent answers you, from the 🤖 button in a Story's header or from your
profile. The picker lists every agent you can use — Verbatim AI's own and yours — each card
showing what it is for and whether it answers Fast or Thorough, with a tick on the
one answering you right now.

Until you pick, that is the agent badged Default, and going back to it is simply picking
its card again — there is no separate "no agent" setting to undo. The choice is yours, not
the Story's: it holds for the conversation you are in and every one after it, and it
survives closing the app. The 🤖 button stays lit while you are on an agent of your own and
names whichever is answering, so you always know what is about to reply. Answers already
written keep the agent that produced them. See Agents.

🤖 Agents: tune how your questions get answered

A new Agents space, in the navigation bar and on your dashboard. An agent is the setup
behind an answer — how widely your documents are searched, which excerpts survive, what
instruction the model receives, how much of the conversation is replayed, and which model
writes the reply. The list shows Verbatim AI's own agents (🔒, read-only, one of them
marked Default) alongside yours, which you can create, tune and delete. Only a name
is required to create one: everything else starts on the platform's settings and follows
them until you pin a value of your own — and clearing a field puts it back to following
them again. See Agents.

📊 See what you have been consuming

A metrics page, in three places. The Statistics card of your dashboard — prompts and
tokens together, where there used to be two cards — opens a report
for your whole organization; a corpus opens its own, from the 📈 icon on its card or
the View Metrics button on its page; and My statistics in your profile opens
yours. Each report gives you the tokens spent, the stories and prompts opened and the
storage taken — as headline totals and as charts over time, cut by day, week, month or
year. See The dashboard.

👁️ Read a document without leaving the app

Preview any indexed document, page by page. Click a file's row in your document list —
or pick Preview in its menu — and it opens in a reader: the page you are on fills the
screen, every page is listed as a thumbnail down the right, and arrows walk you through
the document. Thumbnails load as you scroll, so a two hundred page report opens as
quickly as a one pager. The header gives you the summary of the document and a
download of the original. From a Story, the same reader is one tap from
any source an answer cites. See Documents.

📥 A new way to add documents

Uploading no longer starts the moment you pick a file. The Upload button now opens a
page of its own where you build your selection first: drag and drop your documents
(in the browser and the desktop app), or browse for them — several times over, since
each pick is added to the list rather than replacing it. Remove any row you added by
mistake, send up to 20 files at once, and press Upload files only when the batch
looks right. Cancel backs out at any point, even mid-upload. See
Documents.

🏷️ Tags on your documents

Attach free labels — a team, a year, a project — to the documents you send, and find
them again later. Tags are typed on the upload page and apply to every file of the
batch; up to 32 per document. They then appear under each file's name in your
document list, and Edit in a file's menu lets you change them — and its name — long
after it was sent. See Documents.

⚙️ Advanced upload options

For the cases that need it, an Advanced panel exposes the full ingestion settings:
the chunking configuration (strategy, chunk size, overlap and more), a Provider label
and free metadata. Simple stays the default, and remains the right choice almost
every time. See Documents.

♻️ Replace a document's content

A file that failed to process no longer has to be deleted and re-added. Pick Replace in
the file's menu and send a new version: the document keeps its name and its place in your
Corpus, only the content behind it changes, and your Corpus relearns it straight away. It is
just as handy for pushing an updated version of a document you already rely on. See
Documents.

🔑 API keys for the secured API

Integrate Verbatim AI into your own systems with confidence. You can now publish and
manage your RSA public keys directly in the app: give a key a name and description,
activate or deactivate it, and delete it when it's no longer needed. Your keys are
used to verify the JWT tokens that call the secured Verbatim AI API.

🎫 Instant API token

Grab a fresh JWT token in a single tap — it's copied straight to your clipboard so you
can start calling the API right away.

🆔 Your organisation Id, one tap away

Calling the secured API means telling us which organisation is calling. Your
organization Id now sits in your profile, just under your email — tap the chip to copy
it and drop it into the oid claim of your JWT. See your profile.

---

January 2026

😎 Six AI personalities

Choose the assistant that fits your job. Verbatim AI now ships with six distinct
personalities — Aly (generalist), Léo (relaxed), Nova (impact/marketing),
Lexa (legal), Pia (methodical) and Marcus (tech). Each has its own tone and
expertise. Set a default in your profile and switch it per
Story whenever you like.

🎨 Make the app yours

• Pick your accent color with the new theme color picker.
• Toggle light / dark theme from anywhere.
• Optionally show the build number for easier support.

🖼️ Avatars everywhere

Friendlier profile and corpus avatars give the interface a more personal feel.

---

December 2025

🧭 Sign up in three minutes

A brand-new guided registration: enter your email, confirm it with a verification
code, add your details and accept the terms. Signing up activates a free 3-month trial,
with no commitment. See Getting started.

🔐 Protected registration

Sign-up is now protected by Cloudflare Turnstile to keep your account and our service
safe.

✨ Pro Edition

Discover everything the Pro Edition unlocks: connectors (Google
Drive, SharePoint, Atlassian, Zendesk, Mail and more), Slack & Teams deployment,
team collaboration and API access — all with pay-as-you-go pricing.

📖 Help on every screen

Contextual, richly formatted in-app help is now available throughout the app, so you're
never more than a tap away from guidance.

🎯 Guided Story start

When you open a new Story, a short prompt helps you share your objective in a few words
so your corpus gets the right context from the very first message.

---

November 2025

The first public releases of Verbatim AI, with the core experience.

⭐ AI Corpus

Create and manage themed knowledge bases — one Corpus per domain (Sales, Marketing,
Legal, a project…). See AI Corpus.

📄 Documents

Upload your PDF, Word (.docx) and PowerPoint (.pptx) files, then rename,
download or delete them. These documents fuel your corpus's knowledge. See
Documents.

📚 Stories

Chat with your corpus in natural language and get answers grounded in your own documents,
with the source documents cited. You can also:

• Copy a post or a whole Story to reuse it elsewhere,
• Export a Story to PDF,
• Pin important posts,
• Delete a Story from your history.

See Stories.

🏠 Dashboard

Your home screen shows an overview of your activity — corpus, stories, prompts and token
consumption — plus a Latest prompts list to resume your work instantly. See
The dashboard.

😎 Profile & security

Fill in your professional profile so your corpus knows your context, and protect your
account with a strong password. See your profile.

🌍 English & French

The whole app is available in English and French, with an in-app language toggle.

🌙 Dark mode

Switch to Dark Mode for visual comfort, and back to light mode any time.

🪲 Feedback & bug reporting

Spotted a bug or have an idea? Report it from inside the app — navigate to the screen,
highlight the issue and describe it in a few words.