# Interact with Vera (/docs/tips/interact-with-vera)



**Vera** is the AI assistant built into Ovidens. It is the no-code way to get things
done: you describe what you want in plain language, and Vera does the work, whether
that is building a whole app or making one small change. You will find Vera in the
**Vera (AI)** panel in analytic and indicator app editors.

## What to ask Vera [#what-to-ask-vera]

Vera can handle most of the things you would otherwise build by hand. For example,
you can ask it to:

* **build an app** from a description of your data and your goal,
* **add a widget** to the Visualisation, such as a chart of revenue by month,
* **write a transformation**, so you do not need to know SQL,
* **build a calculation** that works a new metric out from the ones you collect,
* **explain the data**, to help you understand what an app is showing,
* **write or update documentation** in the Knowledge Base,
* **explain how Ovidens works**, with links to the user documentation.

Vera uses the tools enabled for you and allowed on their targets. It chooses the
steps your request needs. If a tool or permission is missing, it says so briefly.

## How to phrase requests well [#how-to-phrase-requests-well]

Vera does its best work when your request is clear and well supported. A few habits
help:

* **Be specific.** "Total revenue by region for this year, as a bar chart" gives
  Vera far more to work with than "make me a report".
* **Give context directly in chat.** Explain what your columns mean and the rules
  you want to use. The [Knowledge Base](../analytical-apps/knowledge-base.mdx) can
  help when relevant, but preparing it is optional. Ask Vera to document a rule if
  you want it saved there.
* **Attach a file** when it helps, for example a sample spreadsheet so Vera learns
  your column names.
* **Answer Vera's questions.** When it pauses to ask for a detail, your answer guides
  the rest of its work.

## Keep the conversation going [#keep-the-conversation-going]

A request is one turn in your conversation. When Vera finishes, send a follow-up
in the same message box, for example "Make that chart blue" or "Explain the last
calculation". Your recent requests, clarification answers and Vera's replies help
it understand what you mean. For a much older topic, restate the important details.

Your conversations are personal to you within the workspace. **Conversation history**
opens a centered table with titles, the initial app context, creation time, last activity
and last opening time. Search by title and use the page controls to browse the archive;
only one page is loaded at a time. Choose **Select** beside a conversation to reopen it.
**New conversation** creates another named thread,
and **Rename conversation** changes its title. Full history is saved; **Load earlier
messages** lets you read older turns.

Vera automatically reopens your last conversation only in the same app and within
6 hours of your last interaction. Opening the panel or background polling does not
extend that window. Otherwise you start with a blank chat; its conversation is saved
when you send the first message. You can manually open any older conversation from
history, including one started in another app. The current page provides context for
your next request. Vera can use tools on other apps when your permissions allow it;
if the target is ambiguous, it asks which app you mean.

An unsent message is kept in this browser tab for the selected conversation,
or for the current app before the first message, separately for each user and workspace. Changing conversations does not cancel a
running task. Choose **Stop this request** if you want to stop it.

Your full conversation history remains available for browsing. When you continue,
Vera uses the latest saved rolling memory and recent messages, including after
a few days. Earlier messages are for browsing; continuing uses the app's current
state and current workspace model. Memory compression can lose detail, so restate
an important fact if Vera no longer has it in context.

Vera condenses older context as it approaches the selected model's available
context capacity, rather than after a fixed number of chat messages. Tool results
also occupy context, so the visible message count alone does not describe its size.
Condensing changes the working memory sent to the model; it does not delete the
saved conversation. The current instructions and authorized tool definitions remain
available on every request. If an older tool-discovery result leaves working memory,
Vera can discover the current tools again.

## How Vera works [#how-vera-works]

Vera shows progress while working. It reads relevant context and uses available
tools according to your request, without a fixed sequence of setup steps. It
asks a focused question if an important detail is ambiguous, then waits for your
answer. Clear requests can proceed without a confirmation round.

Vera finishes with a short answer about what happened or what it could not do.
The **Details** button opens a dialog with a separate AI usage section, timestamps and one steps
timeline. Progress and actual tool results appear together; a failed attempt
shows its operation, target and reason, and a later success stays a separate
attempt. Ordinary progress carries no success claim. Older tool records without
an event time say that the time was not recorded. The dialog does not repeat your
message or Vera's answer.

Requested edits create or reuse the app's draft and follow the same authorization,
audit and versioning rules as manual edits. Asking for help or an explanation
does not create a draft. Review changes before publishing; Vera publishes only
when you ask and the publication tool is enabled for you.

Visualisation saves cannot overwrite a newer shared app state. The editor pauses
while an AI write is active or shared changes are being refreshed. If a refresh
fails, editing stays paused until the retry succeeds. This app-wide coordination
does not expose another person's private conversation.

Vera searches a bundled copy of the public user guides and reads relevant passages
through documentation tools. It has no general internet-browsing tool. The bundled
guides are updated with the application release; Vera does not fetch the live site
when answering. Source links are underlined and open in a new tab, including plain
URLs, so you can read them without leaving your conversation.

## Uploaded files [#uploaded-files]

Files attached to sent requests remain private and currently have no automatic
age-based expiry while their history records exist. Sending or importing a file does
not delete its original bytes. Abandoned uploads that were never sent become eligible
for cleanup after 24 hours. Deleting an app preserves your private conversation
and its source files.

Chat attachment chips currently show the file name and size, without a download
action. A dataset export downloads the imported data, rather than the original
workbook or document. The six-hour conversation-restoration window does not delete
history or uploaded files.

## Task details and version history [#task-details-and-version-history]

Open **Details** beside the response timestamp to inspect tokens, step results and times.
Messages show when requests were sent and responses finished; clarification
questions and answers use their recorded times. Conversation history shows
when the conversation started, when its saved metadata was updated and when you
last opened it. On compact screens, Vera fills the screen; use **Close** to return
to the editor. Resizing back to desktop keeps your saved pane width and draft.

Open the compact **Autonomy** menu beside the attachment button inside the composer to choose how independently
Vera acts with its slider: **0 · Ask first**,
**4 · Balanced** (the default), or **9 · Independent**, with intermediate levels.
You can choose the level before sending the first message. This private conversation
setting applies to new chat requests. Important missing
business details still prompt a question at every level.

The version history shows **Version publisher**, **Version contributors** and
**AI models and clients** directly in both the Versions and Runs tables. The
contributor list records people and AI involved in changes made for that version.
Each model appears with the person who instructed it; an external AI connection
shows its client name because its exact model is not reported. In the Runs table,
these columns describe the linked version, whose number and name appear in the row.
Contributors are not copied from an inherited or restored base.

Version history has no coarse AI icon or link to a private Vera task. The app's
change audit remains available where your permissions allow it. If you publish a
draft that Vera helped edit, the publication is yours. Finishing a Vera request
does not require or force publication.

## Attach original data files [#attach-original-data-files]

You can attach the original CSV (up to 512 MiB) or Excel `.xlsx` file (up to
64 MiB) to a message. Convert older `.xls` files to `.xlsx` first. The upload
remains private until you ask Vera to import selected data into the shared app;
the browser does not recreate the file's rows as chat text.

You can also attach documents, such as a PDF, and ask Vera to read, summarise or
answer questions about them. Reading an attachment does not add it to the knowledge
base or change the app. Ask explicitly if you want information imported or saved.

Pending page and title edits are saved before Vera starts from them. If a save
fails, your request stays in the composer. When someone else changes a saved page
or layout, your editing copy is kept: choose **Load saved** to use their changes,
or explicitly replace the saved copy with yours. Unsaved skill changes stay in
your form; Vera uses the saved skill until you save them.

## Ask for help with Ovidens [#ask-for-help-with-ovidens]

You can ask "How do I upload data?", "How do I publish this app?" or "What does an
indicator dimension do?". Vera can consult the maintained user documentation and
link to the relevant guide. You do not need to copy a help page into your app's
knowledge base to ask about it.

## Model and data processing [#model-and-data-processing]

Vera uses Anthropic's API directly. A workspace administrator chooses one model:
Claude Haiku 5.5, Claude Sonnet 5.5 (the default), or Claude Opus 5.5. Every chat,
AI widget and internal AI task uses that workspace model. New and resumed requests
use the current setting, including in older conversations.
**AI processing is not limited to the EU.** Your request, relevant
conversation context, attachments and tool results may reach the provider. See
[AI transparency](../infrastructure/ai-transparency.mdx) for processing details.

## AI widgets are read-only [#ai-widgets-are-read-only]

An [AI widget](../analytical-apps/skills.mdx) runs a reusable task against the
app's existing configuration and published data. It returns an answer without
changing data or configuration, running the pipeline, publishing, or creating a
draft, even if the person using it has permission to edit the app. A preview can
read an existing draft skill without changing it. A widget request needing a
change receives a short explanation that it cannot make that change.

Widget answers are separate from your conversation in the Vera panel.

## When Vera cannot finish [#when-vera-cannot-finish]

Vera gives a brief reply explaining what it could do or what stopped it. If a
closing reply is unavailable, a short explanation appears instead. Open **Details**
for recorded tool attempts, their reasons, token usage and times. Earlier failed
attempts remain visible even when a later attempt succeeds and the task completes.

An **Internal AI request error** means the provider could not accept the request
Vera prepared. Try again; contact Ovidens support if it keeps happening. This is
reported separately from unavailable AI connections, missing configuration or a
busy provider. **Details** remains available even if the request failed before its
first tool call. If token usage was not recorded, the dialog says so; its counters
do not establish the provider's full usage.

Continue in the normal message box to supply a missing detail, rephrase a request
or try again. Requested edits remain in the app's draft until you publish or discard
them; review those changes through the app's own controls.

### If your connection drops [#if-your-connection-drops]

A lost connection does not stop Vera: the request keeps running on the Ovidens side.
If the panel cannot reach Vera for a moment, it says **Lost connection to Vera -
retrying** and keeps trying. If it still cannot reach Vera after a while, it says
**Couldn't reach Vera. The run may still be going.** and offers a **Check again**
button, which picks the request up wherever it has got to.

### When the Vera panel is closed [#when-the-vera-panel-is-closed]

In an analytic app's editor, the **Vera** button shows a small dot while the panel is
closed: blue while Vera is working, amber when Vera is waiting for your answer, and
red when a request ended without doing everything you asked. The red dot stays until
you open the panel.

## Vera knows the system, not your use case [#vera-knows-the-system-not-your-use-case]

Vera understands Ovidens inside out: how pipelines, indicators, and the
Visualisation work, and how to build them. What it does **not** automatically know
is your business, your data, and what you are actually trying to achieve. Give it
that context and it can build very quickly and very well; leave it out and it has
to ask you for missing details.

So before you send a request, ask yourself whether you have told Vera everything it
needs from a business point of view: what the numbers mean, which rules apply, and
what a good result looks like. Putting that context in your request, or in the
[Knowledge Base](../analytical-apps/knowledge-base.mdx) when useful, is the
single biggest thing you can do to get a better result.

And like any AI, **Vera can make mistakes**. Everything it builds lands in your
app's draft, which viewers do not see until it is published, so always check its work
before you rely on it, especially the logic behind an important number.

## A note on AI usage [#a-note-on-ai-usage]

Working with Vera uses **AI credits**, which your workspace has a monthly allowance
of. You can see how much you have used on the [Administration](../administration.mdx#usage)
page. If the monthly limit is reached, AI features pause until the allowance resets
at the start of the next month, while the rest of the app keeps working as normal.

Vera requests five-minute prompt caching for reusable input. Cache reuse depends
on the provider, an unchanged input prefix and the cache lifetime; saved chat
history is separate from that temporary cache. Condensing memory can require a
new cache write and another model call, so a shorter context is not always cheaper.