You can reach the platform at chat.novendix.AI. You’ll receive your login details from the person in your company who manages the licenses – each license is assigned to exactly one named person.

After logging in, you land directly in the chat. There’s no installation, no client, and no plug-in: the platform runs entirely in the browser, including on tablet and smartphone.

There are three areas:

• Sidebar on the left – can be expanded and collapsed via the icon top left. It contains New Chat, Files, Skills, Agents, Projects, Tasks, Workflows, Meetings, App Connections, MCP Servers and Memory. Below that is the search over your history, the chat history grouped by time period, and your name at the very bottom.
• Header – select an agent, select a model, and the Private / Public toggle for the current chat.
• Input field at the bottom – type text, attach a file (paperclip), template library (book), toggle web search on and off, dictate (microphone), and the indicator showing how full the conversation already is.

Three things almost always help: say who you are or who the result is for, say what the outcome should be (bullet points, email, table, two sentences), and provide material, instead of describing it.

So instead of “Write something about data protection,” try: “I’m the IT manager at a trade business with 40 employees. Write a short staff notice explaining why, from now on, we’re only using this platform for AI. Maximum 200 words, factual tone.”

Above the input field are four suggestions you can tap if you’re not sure how to start.

There are three packages. They build on each other:

• Starter – chat with the standard models, Smart Select, use ready-made agents from the store, your own skills, projects, files, and memory.
• Business – in addition, all model tiers (Standard, Premium, Ultra), create and publish your own agents, share chats and agents via link, plus app connections to over 3,000 tools.
• Enterprise – in addition, your own MCP servers, custom system integration, and priority support.

The person in your company who manages the licenses knows which package is assigned to you. Anything your package doesn’t include is either not visible or not selectable in the interface.

The platform has been completely redeveloped. The chat itself works as before, but a lot has been added around it:

• Documents – the AI creates finished Word, Excel, PowerPoint, and PDF files, edits existing PDFs, and fills in Word templates.
• Canvas – longer texts are created in their own workspace next to the chat, with versions, undo, and comments.
• Tasks and Workflows – recurring jobs on a schedule and multi-step processes with triggers, branches, and outputs.
• Meetings – recording, transcription, and summary.
• Skills and skill packages, Agent Store, Memory as its own area and a full-text search across all files.

Some individual features of the old interface no longer exist in this form. If you’re missing something, let us know – we’ll look into it.

By default, Enter sends the message, Shift + Enter inserts a line break. If you’d rather work the other way around, this can be changed under Settings › Chat › Enter to Send .

Hover your mouse over your own message – a pencil and a copy icon appear below it. Use the pencil to change your question and have it answered again from that point.

This is almost always better than appending the question again at the bottom: the history stays short, and the AI isn’t distracted by your first, imprecise attempt.

Hovering over any response reveals three icons: Copy, Thumbs up and Thumbs down.

Rating isn’t just a formality – it helps us see where model selection or a skill needs adjusting.

To the right of the microphone is a small circle. Clicking it shows how much of the context window is in use – for example “35 / 200,000 tokens,” broken down by input and output.

The context window is the memory of a single chat. Everything in it is carried along with every new question. That’s why very long conversations become slower and more expensive.

Rule of thumb: one topic, one chat. When you change topics, start over via New Chat .

If a long conversation still needs to continue, have the current state summarized in five bullet points, copy them into a new chat, and continue there. Permanently important facts don’t belong in the chat – they belong in Memory or in a Project.

In the sidebar, above the history, is the field Search history…. The history itself is grouped by time period – Yesterday, Last 7 days and so on – and can be expanded.

You won’t find files here, but in the Filesarea. That also searches the content of documents, not just the file names.

In the history, click the three dots next to the chat, then Delete. This removes the conversation along with its history.

Files created in this chat remain in the Files area and must be deleted separately there. There is no trash/recycle function – deleted is deleted.

At the top of the header is the Privatetoggle. Set it to Public, and anyone with the link has access to this chat. The same toggle is also in the chat’s three-dot menu in the history, under Share.

Please check beforehand what’s in the conversation. A public link is not password-protected – whoever has it can see the whole chat. Switching back to Private ends access.

Three dots next to the chat in the history, then Add to Project and select the project. The chat then also appears in the project view.

The book icon in the input field opens your Templates – saved phrasings you need again and again. Use the plus to create a new one. Templates can be private or apply to the whole workspace.

Rule of thumb: a Template is a text snippet you insert deliberately. A Skill takes effect automatically when it fits the situation. An Agent is its own counterpart with a model, knowledge, and tools.

Smart Select is the default setting. It looks at your request and automatically selects a suitable model for it – a fast one for short questions, a more powerful one for analysis or code.

For everyday use, this is the right setting. You don’t need to remember model names or figure out which one is currently best.

There’s also GPT Image Smart Select – the same automation, but geared toward image generation.

The models are sorted into three tiers:

• Standard – fast and inexpensive, for most everyday work.
• Premium – noticeably stronger at analysis, longer texts, and code.
• Ultra – the most powerful models, for truly demanding tasks.

In the Starter package, the standard models are available. From Business onward, all three tiers are unlocked.

Clicking the model name in the header opens the list with a search field. Currently, alongside Smart Select and GPT Image Smart Select, it includes:

• Standard: Claude Haiku 4.5, Gemini 3.5 Flash Lite, Gemini 2.5 Flash, Gemini 2.5 Flash Lite, GPT 5.6 Luna, GPT-5 Mini, GPT-5 Nano, o4-mini, Mistral Small, Devstral
• Premium: Claude Sonnet 5, Gemini 3.8 Flash, Gemini 2.5 Pro, GPT 5.6 Terra, GPT-5, Mistral Large, Mistral Medium 3.5
• Ultra: Claude Opus 5.5, Claude Fable 5.1, GPT 5.6 Sol

The list is continually expanded; the menu in the platform is always the authoritative source.

The wrench means “supports tools” – the model can use web search, file creation, connected apps, and MCP servers. The eye means “supports image recognition” – the model can view images and scanned pages.

If you upload an image and don’t get a useful response, first check whether the selected model has an eye icon.

It’s worth it in three cases:

• Very long or complex texts – an Ultra model delivers noticeably better structure here.
• Code – Devstral and the larger models are better suited for this.
• Viewing images – choose a model with an eye icon.

For everything else, Smart Select is the better choice, because it doesn’t use unnecessarily expensive models.

This works at any time: select a model at the top, keep writing. The previous history is preserved and carries over to the new model.

A typical flow: research with Smart Select, then switch to a Premium or Ultra model for the final version.

Via your name at the bottom left › Book Budget you can book credit packages – from €20 to €1,000. Each package shows how much credit and roughly how many tokens it contains.

As the recipient, choose the team pool or an individual member. Payment isn’t made by credit card in the tool: the booking appears collectively on your next regular invoice, in accordance with your existing contract.

Click the paperclip in the input field, select a file – multiple files at once are possible. The content is made available as text so you can refer to it in the chat: “Summarize pages 3 to 5,” “What deadlines are in the contract?”

Directly supported are PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), CSV, JPG, PNG as well as the audio formats MP3 and WAV.

Other formats aren’t read automatically. For an unusual format, it usually helps to copy the content as text into the chat or save the file as a PDF first.

In the sidebar under Files all documents are in one place – both the ones you uploaded and the ones the AI created. For each file, you can see size and date, and you can download or delete it.

The search field in the Files area searches file name and content. So you’ll also find a document whose name you’ve forgotten, if you still remember a word from its text.

In everyday use, this is often the fastest way to the right document – faster than any folder structure.

Just tell the chat what you need: “Create a Word document about …,” “Build me an Excel spreadsheet with …,” “Turn this into a PowerPoint with eight slides.”

The finished file appears as a card directly in the response, with a download icon – and is also available under Files. Excel spreadsheets contain real formulas, not pre-calculated numbers.

Upload the PDF and tell the chat what should be changed – for example, “change the price to €149.” The original layout is preserved, only the affected value is replaced.

Upload your letterhead template as a Word file and describe what should go into it: salutation, date, text sections, tables. The template’s layout remains unchanged.

For mail merges and recurring letters, this saves the most effort.

Describe the image in the chat – “Generate a minimalist icon for ….” The finished image appears in the response, can be viewed larger with a click, and is then available under Files.

The more specific the description, the better: subject, style, colors, background. If it doesn’t fit, describe the change and have it regenerated.

To the right of each file are a download icon and a trash icon. Deleted files cannot be restored.

Please download anything you need permanently and store it in your own filing system. The platform is not an archive and not a substitute for your data backup.

When a longer text is created, a separate workspace opens next to the chat – the Canvas. The text sits there large and coherent, instead of disappearing into the chat column.

You can also explicitly request the Canvas: “Please write that in the Canvas.”

You have two options: either write in the chat what should be changed – then the AI updates the document – or edit the text directly yourself via the pencil icon.

Above the text, it shows when it was last updated.

On the right edge of the Canvas are the tools: Version history, Undo and Redo, Copy, Download, Comment and Edit.

If a revision made the text worse, you can go back via the version history – you don’t have to reconstruct anything by hand.

The X at the top closes the area; the chat remains intact. It can be reopened via the corresponding response in the chat. Whatever you want to keep using, download it or have it output as a Word document.

A Skill is a reusable instruction that supplements the AI’s behavior in the chat – can be switched on and off individually. It takes effect automatically when the situation fits.

Example: the skill Excel Spreadsheets ensures that every generated spreadsheet contains real formulas, clean formatting, and no formula errors – without you having to specify that every time.

Under Skills › Skill Packages are thematically bundled collections:

• Documents – create and edit Word, Excel, PowerPoint, and PDF.
• Compliance & Operations – GDPR review and structured incident response.
• Engineering – look up Git commands and regex patterns.
• Platform – help using novendix.AI, writing your own skills, onboarding for new users.

For each package, it shows how many of the included skills you’ve already installed.

Under Skills › All you see everything available and install with one click. Under Installed are your active skills, each marked Active as well as a pencil and trash icon.

Less is more: install what you really need. Every active skill affects every conversation.

At the top right, via Create Skill. What matters most is the description of when the skill should apply – it controls in which situations it takes effect. Good descriptions start with “Use this skill when ….”

If you get stuck: the built-in Skill Creator guides you in the chat from a vague idea to a finished skill.

Above the response, you’ll then see the note SKILL ACTIVE. If a skill is getting in the way, switch it off under Skills › Installed .

A Agent is a configurable counterpart with its own instructions, a fixed model, its own knowledge, and its own tools. You select it at the top in the header and then chat with it instead of the general chat.

This makes sense anywhere the same role is needed repeatedly – proposal texts, support responses, review against an internal policy.

Under Agents › Browse Store are publicly shared agents from other users – for example, Email Assistant, Copywriter, IT Security Advisor, Digital Transformation Coach, or GDPR Reviewer.

With Copy to My Agents you get your own copy that you can customize. The original remains unchanged.

Via Create Agent. The most important fields:

• Name, icon, description – so colleagues can recognize what it’s meant for.
• Instructions – the actual behavior, up to 50,000 characters.
• Model and creativity – fixed model or Smart Select, plus a slider from factual to creative.
• Conversation starters and tags – starter questions and categorization.

Four blocks have proven effective: Role (who you are), Task (what you do), Rules (what always applies, what never does), Format (what the response looks like).

Write in full sentences and state limits explicitly, for example: “If information is missing, ask rather than making it up.”

In the Knowledge section, you attach files to the agent. They are searched, and matching excerpts are incorporated into the response as needed.

This way you give an agent your price list, your policy, or your manual, without uploading it again in every chat.

Per agent, you can enable Web search, and image generation and Canvas .

There’s also an agent-specific memory: the agent permanently remembers relevant facts from conversations with you. It applies only to this agent – your general memory is independent of it and remains active in any case.

You can give the agent fixed fields. When starting a new chat, these fields then appear instead of the free text field.

For colleagues who don’t want to compose text, this is the simplest form: fill in, submit, done.

Under With Evals you create test cases. They automatically check whether the agent meets a criterion – a second AI evaluates the response.

This is worthwhile before you roll out an agent to the whole company, and after any major change to the instructions.

Under With custom APIs an agent can call its own HTTP endpoints. This lets you connect specialized systems for which no ready-made app connection exists.

Please coordinate this with your IT department – an endpoint the AI is allowed to call is a deliberate decision.

Three levels: Private – only for you. Workspace – for your organization. Public – in the store, copyable by everyone.

Before publishing: check whether the instructions or knowledge contain internal information. Public really does mean public.

A Project bundles multiple chats under shared files, instructions, and automatic memory. Under Projects › Create Project you assign an icon, name, description, and visibility (private, workspace, shared).

Typical: one project per customer, per tender, per internal initiative.

Two ways: in the project view, click New Chat in This Project – or, for an existing chat in the history, click the three dots and Add to Project.

The project view then lists all associated conversations.

Under Memory contains all the memories your agents and projects have stored about you – sorted chronologically and labeled by category.

The Memory for Chats toggle controls whether the platform also automatically remembers permanently relevant facts from regular chats.

Check it from time to time. Delete anything that’s no longer accurate – otherwise the AI keeps working with outdated assumptions.

And the other way around: anything that shouldn’t be here doesn’t belong in the chat either. Health data, HR matters, or login credentials have no place in a chat window.

At the top right of the Memory area are Import and Export. This lets you back up your memories or carry them over into another account.

A Task is an instruction to the AI that runs automatically at set times. Each run appears as a new chat in your chat list.

Examples: preparing the weekly overview every Monday, summarizing the most important industry news every morning.

Under Tasks › Create Task you specify:

• Prompt – what the AI should do.
• Name – for recognizing it in the list.
• Trigger – once, hourly, daily, on weekdays, weekly, or advanced.
• Run at and Time zone – preset to Europe/Berlin.

Write the prompt as if someone with no prior knowledge were reading it – when it runs, there’s no conversation history the AI can refer to.

Under Under Advanced you also select the AI model and optionally a Agents or an Project as additional context.

A A workflow is a multi-step process with triggers, AI steps, and outputs that you assemble on a canvas. While a task runs a single job at a set time, a workflow chains together multiple steps and can branch.

Under Workflows › Templates are ready-made processes, sorted by Marketing, Support, Development, Sales, and Own – for example, “Draft a proposal,” “Categorize incoming email,” “Route support ticket by priority,” or “Prepare weekly newsletter.”

With Use Template you get your own copy that you can customize. This is the fastest way to get started.

Via Add Node:

• Trigger – manually, on a schedule, or via webhook when an external URL is called.
• AI – AI Processing (generate content) and Web & Research (search and read the web).
• Control – Condition (If/Else) and Distributor (multiple named branches).
• Outputs – send to chat, send email, push notification, call webhook.
• Other – note on the canvas.

At the top right are Save and Run Now. Test every new workflow manually first, before activating a schedule or webhook.

Results sent back to you collect in the Inbox tab.

In the input field is the Web searchtoggle. It is active by default. As long as it’s on, the AI can retrieve current pages instead of only answering from its training data.

Turning it off is worthwhile if you’re working purely with your own documents and don’t want external sources in the result.

If a search was performed, an expandable step appears above the response: Web Search – Completed. It shows, under Sources , the pages used, with title and domain.

For anything that goes out externally or is legally relevant, please expand this and check the sources yourself.

The microphone on the right of the input field starts voice input. The first time, the browser asks for microphone permission.

It works best if you speak at a normal pace and in complete sentences. The result can still be corrected before sending.

The Meetings area provides AI-powered meeting notes with transcription and summary. Via Start Meeting the recording begins; the transcript and summary are then available in the list.

Recording a meeting is not a legal detail. Obtain the consent of everyone involved beforehand and state explicitly that it is being recorded and automatically transcribed.

For works council matters, HR conversations, and anything involving employee data, please coordinate this beforehand with the responsible department in your company.

Under App Connections you connect third-party apps – Gmail, Outlook, Slack, and many more – so the AI can use them as tools. Find the app in the catalog and click Connect.

. Login takes place with the respective provider. Only grant the permissions that are actually needed.

You can then tell the chat what should happen – for example, create an appointment, draft a message, or retrieve data from a system.

Treat write actions like a signature: look them over once before they go out. The AI carries out what you instruct and doesn’t check whether that was really your intention.

In the App Connections area, at the relevant app. It’s also worth revoking access with the provider itself – under their security or app settings.

Via MCP Servers in the sidebar, external tool servers can be connected using the Model Context Protocol. Active servers are available to the AI as additional tools.

You’ll need a name, the server URL, the transport type (Streamable HTTP or SSE), and optionally a bearer token. This setup belongs in the hands of your IT department – a connected server can trigger real actions in your systems.

Via your name at the bottom left › Settings. There you change your display name and profile picture and see your registered email address.

Under Appearance you choose between Light and Dark.

Also in Settings, under Chat. If the toggle is on, Enter sends and Shift+Enter makes a line break. If it’s off, it’s the other way around.

Settings › Data › Exporting all chats. You receive your conversations as a file to download.

If you need content permanently, this is the right way – together with downloading the files from the Files.

Also under Dataarea. This removes your entire chat history and cannot be undone. Export beforehand if you want to keep anything.

Via your name at the bottom left › Logging out. On unfamiliar or shared devices, please always log out instead of just closing the tab.

The platform runs in a German data center. The AI models are accessed via an upstream provider within the EU; only models without training on your data and without storage at the model provider are approved.

The details – legal bases, subprocessors, technical and organizational measures – are on the Privacy & Security and Data Processing Agreement.

No platform can take three things off your hands:

• What you put in. Special categories of personal data – health, religion, union membership, sex life – don’t belong in a chat. Nor do login credentials and keys.
• What you take out. Check results before they go out. AI can be convincingly wrong.
• Your own backup. Download whatever you need permanently. The platform is not a substitute for an archive or a data backup.

In this order:

1. Edit the question instead of appending a new one – via the pencil on your message.
2. Be more specific – state audience, length, tone, format.
3. provide material instead of describing it.
4. Switch models – go one tier higher for demanding tasks.
5. Start a new chat – if the conversation has gotten stuck.

Most common causes: the format isn’t read automatically (PDF, Word, Excel, PowerPoint, CSV, JPG, PNG, MP3, and WAV are supported), the PDF is a pure scan with no text layer, or the selected model doesn’t have an eye icon for image recognition.

Fix: choose a model with an eye icon, save the file as a searchable PDF, or copy the text portion directly into the chat.

Then it belongs to a tier your package doesn’t include. Premium and Ultra are unlocked starting with the Business package. Speak to the person in your company who manages the licenses.

Usually the authorization with the provider has expired or the password was changed. Disconnecting and reconnecting resolves this in most cases. If the problem persists, contact us – with the name of the app and the approximate time.

Write to info@novendix.ai or call: 09122 98228-0.

Three pieces of information help us most: what you were trying to achieve, what you entered, and what came out. A screenshot usually saves a follow-up question.