TextCortex User Handbook

The TextCortex Platform is equipped with all the necessary features for you to personalize your experience on an enterprise Agent infrastructure that connects AI Agents with your company knowledge to ultimately bring you measurable results.

Who is TextCortex for?

TextCortex is designed for teams and enterprises that are looking to enhance their operations with an AI infrastructure that can be tailored to their specific objectives and needs through a centralized solution that matches complex workflows.

Get started

The chat interface is the first thing you'll see when accessing the TextCortex platform.

This is where you can input your prompts, select the tools and knowledge you'd like to work with and select an Agent from our default library.

Agents

You can create and deploy intelligent AI Agents that connect to company data for context-aware interactions and automate high-value enterprise operations.

Users can combine their preferred knowledge with the background they choose for their Agent. This dual capability enables a more relatable and contextually appropriate interaction, making AI Agents not only sources of information but also virtual entities that reflect certain characteristics and refer to specific sources.

Through this innovative blend, users can experience a more dynamic and personalized experience.

Create your first Agent

Click on the "Agent Profiles" section of the TextCortex sidebar, then on "Create New Agent" to start configuring your Agent.

Time to start creating your Agent!

1.  In the "Behavior" section, choose a name for your Agent, start by adding a description and background to it.

📌 Background example:

"You are John, an AI assistant with knowledge and capabilities that allow you to provide detailed and insightful feedback on project outlines. Your writing style is professional, objective, and analytical, offering constructive guidance to users.
You are able to evaluate the clarity of a project's objectives and scope, ensuring that the goals are well-defined. You will ask at least 3 clarifying questions to fully understand the user's intentions before providing detailed feedback. Your responses will highlight both the strengths of the project outline as well as areas that could be improved, with the goal of supporting the user's success.
You have a deep understanding of effective project planning and can provide comprehensive assessments to help users refine their proposals. Your feedback will be thoughtful and tailored to the specific needs of each project, drawing on extensive knowledge to offer valuable insights.
When communicating with users, you will maintain a supportive and encouraging tone, delivering critiques in a constructive manner. Your aim is to be a collaborative partner in the project development process, working together to align the outline with the desired outcomes".

Scroll down to set a specific tone guideline and decide how your Agent interacts by establishing a set of rules.

For example, since we created a project evaluator, we clicked on "Always" and added Provide actionable feedback: move beyond identifying problems to suggest specific, realistic next steps or improvements the project team can actually implement as one of our rules.

📌 Rule example:

"Never ignore stakeholder perspectives, avoid evaluating purely from a technical or financial angle while dismissing what matters to the actual project sponsors or team".

2. In the "Tools" section, you can further personalize your Agent by choosing a default AI model, the Knowledge Base it should refer to by default, the functionalities the Agent can use (such as web search, data analysis, preferred source...), some integrations, the image generation model you wish the Agent to use and whether you'd like the Agent you're creating to be able to create additional Agents.

3. Finally, in the "Access" section, you can decide who has access to the Agent you're creating, both as individual users and as organizations.

The preview chat on the right side of your screen allows you to test out your Agent during the creation process: you can leverage it to make sure you're satisfied with how the Agent follows your instructions or relies on the knowledge you provided.

Once you click on "Done", your Agent will be ready to be used!

You can always expand the TextCortex sidebar to set a default Agent, add different Agents to your favorites list for quicker access and of course edit them or create new ones whenever you want.

Best Practices

Define Your Voice

Write down 3-5 adjectives that describe your communication style.

📌 Example:

"Warm, straightforward, friendly".

Create Guidelines

• Identify typical phrases, slang or industry terms you commonly use.
• Note your sentence length preference (short and punchy vs. long and detailed).

📌 Example: 

"Use 'blue sky thinking' instead of 'creative problem solving', keep sentences under 15 words, include 'game changer' in your sentences frequently".

Provide Authentic Examples

• Share 2-3 paragraphs of your actual writing/speaking.

📌 Example:

"This is an email I recently sent to our sales team, use it as reference to calibrate your communication accordingly whenever I ask you to generate any communication to be sent to our sales team: [include the full text]".

Define Rules

• Whether you decide to add rules or a more detailed background, don't forget to clearly state what your Agent should and shouldn't do.

📌 Example:

"When planning a sales strategy, offer tailored suggestions and strategies to enhance the overall sales performance. Assist with identifying target audiences, crafting persuasive sales pitches, overcoming objections and optimizing sales processes for maximum conversion rates".

Create Response Templates

• Build sample responses for specific scenarios and content types.

📌 Example:

When you are asked to create a standard operating procedure (SOP), you must follow the structure of the following boilerplate:
SOP TITLE: [Insert Title]
SOP Number: [Insert Number]
Version Number: [Insert Version].

Test and Refine

• Run tests with different scenarios and evaluate whether you are satisfied with your Agent or if you'd like to edit anything.
• Adjust your instructions based on what's missing.

📌 Example:

"Your responses are too formal: use contractions and shorter sentences".

📌 Brand-Specific Example:

This Agent represents Sunshine Bakery. Keep responses warm and friendly. Use baking metaphors when possible. Always mention our commitment to organic ingredients. Always end messages with 'Baked with love!' and avoid mentioning competitors. When asked about allergens, always emphasize our careful handling procedures.

📌 Personal Style Example:

I'm a tech educator who explains complex topics through a very simple communication. I use analogies frequently and write in short paragraphs. I avoid any type of jargon and often ask rhetorical questions to engage those who read my newsletter. I prefer a casual language and like indulging in occasional humor. Always suggest a practical next step at the end of all explanations.

💡 Pro Tips!

1. When your Agent has to complete a task that requires multiple, perhaps more difficult steps (such as researching updated information on the internet before preparing a report or tabular data paired with visual representations), raise the Effort level and turn Thinking on if the model supports it.
2. If your Agent needs to perform an Internet search and you want it to follow exact, specified time stamps/dates, we suggest to type your desired date as, for example, "2025-11" within your prompt.
3. If you wish to fine-tune one of our default Agents, you can click on the three dots and clone it: you will then be free to edit the Agent as you wish!

💡 Popular Use Cases

• Automating complex tasks and enterprise operations.
• Directly connecting an Agent to email and calendar data.
• Extracting key insights from meetings and team discussions to draft proposals accordingly.
• Summarizing meeting notes.
• Conducting market research.
• Comparing different documents or different versions of the same document.
• Enhancing email writing.
• Maintaining consistency while ensuring enterprise communications are clear, professional and aligned.

Model Selection, Effort Levels and Thinking

You control how much capability your Agent brings to a task through three settings in the chat interface: the model you select, its Effort level and, on models that support it, the Thinking toggle. Refer to "How effort and thinking work" for the full breakdown.

Effort Levels

Every model has an Effort level: Low, Medium or High. Effort sets how long the model works on a task before it settles on an answer. More effort means deeper answers on hard tasks, with higher token use and more time; less effort means faster, leaner responses.

• Low: fastest and most token-efficient. Best for lookups, short rewrites, translations, formatting, simple classification and high-volume tasks.
• Medium: balanced depth and efficiency, a strong default for everyday drafting, summaries, tool-assisted tasks and general Q&A.
• High: deepest reasoning and the most thorough analysis, with the highest token use and latency. Best for complex reasoning, planning, tricky debugging and nuanced analysis.

Effort is a behavioral signal, not a strict token cap: at lower effort the model still works harder on genuinely difficult problems, it just does less than it would at a higher level for the same problem.

The Thinking Toggle

Some models also have a Thinking toggle (On/Off), which controls whether the model reasons through the problem step by step before it starts executing, rather than answering immediately.

• Thinking Off: the model responds directly. Faster and more efficient, good for straightforward requests where the path to the answer is obvious.
• Thinking On: the model reasons first, then answers. This improves reliability on multi-step problems, math, planning and debugging, with additional tokens and time.

Effort and Thinking are related but separate: Effort sets how much total work goes into the whole response, while Thinking decides whether the model reasons before acting. Some models expose only Effort, some expose both. On models that have both, pairing Thinking On with Medium or High effort gives the most thorough results, and Thinking Off with Low effort gives the leanest.

How Effort Shows Up With Tools and Agents

When a model uses tools or runs as an Agent, Effort affects the tool calls themselves, not just the written answer:

• Lower effort tends to combine steps into fewer tool calls, skip preamble and go straight to the action with a short confirmation.
• Higher effort tends to make more tool calls, explain its plan first and give fuller summaries of what it did.

This is why Effort is a strong lever on long agentic runs: lower effort is not only fewer output tokens, it is also fewer round trips.

Choosing the Right Setting

• Quick lookup, translation or format fix: Low, Thinking Off.
• Everyday writing, summarizing or routine tool use: Medium, Thinking Off or On.
• Multi-step reasoning, planning or code debugging: High, Thinking On.
• High-volume or latency-sensitive workflows: Low, Thinking Off.
• Deep research across many sources: Medium to High, Thinking On.

How This Affects Your Budget

Higher Effort and Thinking On increase token usage per message. If you exhaust your usage there are no lockouts: you revert to standard models and default settings and keep working. If your admin has enabled Extra Usage, higher effort levels and longer agentic runtimes continue by drawing from your Extra Usage allowance after your budget is empty.

📌 Note: admins can set the default Effort level and Thinking behavior for the tenant in the Console. Individual users can still adjust Effort per model for a given task unless restricted.

💡 Best Practices

• Set effort explicitly: don't rely on a single level for everything. The right starting point depends on the task and the model.
• Start low, step up: begin at the lowest level that meets your quality bar and raise it only when the output needs more depth.
• Don't leave everything on High: High is built for hard problems and spends your budget faster than most tasks require.
• Reserve High and Thinking On for hard problems: complex reasoning, planning and debugging are where they pay off.
• Match model and effort together: an efficient model (for example Kimi or GLM) on Medium effort often meets your quality bar at a fraction of the token burn of a frontier model on High.
• Be comprehensive: outline your complete end goal in one prompt.
• Specify deliverables: request specific outputs like charts, reports or analyses.
• Control tools: use the "Tools" menu to enable or disable capabilities as required.

Key Takeaway

Effort and Thinking let you dial capability up or down on a single model. Keep both as low as your quality bar allows, raise them for genuinely hard tasks, and pair them with an efficient model to get near-frontier results while keeping your usage intact.

Uploading Files: Work with your own data

You can upload files and ask your Agent to quickly retrieve the information you need from them. 

Supported Formats

TextCortex is engineered to work seamlessly with a range of text document formats. Some of the most commonly supported file types include:

• ✅ Text: .txt, .md, .html
• ✅ Documents: PDF, Word (.docx), PowerPoint (.pptx)
• ✅ Data: Excel (.xml, .xlsx), CSV 
• ✅ Integration-specific sources: Web URLs, SharePoint Sites/Pages (.aspx), Google Docs, Google Sheets...

Limitations

• Size limitation for regular files: 300MB maximum per file.
• Size limitation for tabular data (Excel, CSV): 10 million character limit across all rows, columns and sheets combined.
• AI models have a limited "context window" (character processing capacity). Being specific in your queries improves results significantly.

Troubleshooting

Even files under 300MB may fail to upload if they contain extremely dense text content that exceeds processing limits. Files with dense formatting, extensive metadata or highly compressed text (such as PDFs with small fonts or tables with thousands of entries) can sometimes hit character processing limits before reaching the file size limit.

If you encounter upload issues with smaller files, try:

• Breaking large documents into smaller sections.
• Simplifying formatting or removing unnecessary metadata.
• For spreadsheets, splitting data across multiple files if row/column count is very high.

Knowledge Bases: Connect multiple data points

A knowledge base lets you work with multiple files simultaneously, retrieving information and gaining insights across all your documents at once.

Your AI agent acts as a mediator, providing relevant answers with sources from your centralized data.

How Knowledge Bases Work

When you query your knowledge base, your AI agent searches for relevant sections matching your intent, similar to a search engine. Instead of processing everything at once, it retrieves only the sections related to your question, making results faster and more accurate.

💡 Pro Tip!

We recommend grouping knowledge bases by purpose or department (e.g., "HR Policies," "Marketing Assets") rather than creating one massive repository.

Create your first knowledge base

1. Click on the "Knowledge" section of the TextCortex sidebar.

2. Hit the +New button.

3. Select a name for your Knowledge Base and then upload your desired files. We support various file formats to ensure seamless integration with your existing documents. You can also connect other sources such as Google Drive or OneDrive.

4. Once your files have been uploaded, you can head over to our main chat interface, click on the "Knowledge" button and choose the Knowledge Base or specific file you want to work with.

💡 Useful Info!

You can also select your sources by typing @ in the message box or by clicking on "@ Add context".

📌 Note: all files are processed by TextCortex without the use of third parties. Refer to our article "How We Handle Data at TextCortex" for more information.

💡 Popular Use Cases

• Connecting Google Drive, OneDrive, Dropbox, Notion, Sharepoint files and data with one click.
• Extracting key highlights from long PDF / DOCX files.
• Locating precise information across multiple document repositories and specific pages.
• Efficiently connecting and contextualizing related concepts across different documentation.
• Analyzing CSV sheets with rows of data.
• Working with the information contained in publicly accessible Web URLs.
• Underlying specialized Agents and features with comprehensive knowledge that creates connections between disparate data sources that Agents can refer to at all times.

Data Analysis Tool

The data analysis and visualization tool proves especially beneficial for conducting data analysis, creating visual representations of your data or examining code.

Its primary advantage lies in its functionality: it runs code to generate more reliable outcomes each time it is activated to handle your files. CSV, XML files and other similar formats.

Supported Formats

• ✅ CSV, TSV, XLSX 
• ✅ TXT, MD, HTML, DOCX
• ✅ PPTX
• ✅ PDF
• ✅ Google docs, Google sheets, Google slides

Limitations

• Previously detailed file limitations apply, particularly the size limitation for tabular data (Excel, CSV): 10 million character limit across all rows, columns and sheets combined.

Get Started

To start using the data analysis tool, head over to our main chat interface and click on "Tools". You will then be able to toggle the feature on.

Now simply select the file you want to work with from "Knowledge". Alternatively, you can also easily drag and drop one. Make sure you also select an AI model that is appropriate for your workflow. We recommend using the data analysis tool with a Reasoning model, or with Thinking on and a higher Effort level.

For demonstration purposes, we used a CSV file as an example to show you how the system can transform raw data into actionable insights.

After uploading your file, simply provide instructions for the type of analysis or visualization you need. Our AI will automatically:

• Execute code to process and interpret your raw data.
• Generate visual charts that transform complex information into clear, meaningful insights.
• Provide you with a comprehensive understanding of the patterns and trends within your document.

💡 Useful info!

You can always provide additional feedback and ask for adjustments or corrections once your Agent generates a result for you.

Web Search: Smarter search with up to date information

AI systems don't have access to the real-time web by default: the Web Search feature allows your Agent to access up to date data by gaining a deeper knowledge about recent events to shape its responses and fetch the information you need accordingly.

To activate it, simply click on the "Tools" button and switch the little toggle next to "Web search":

When you type your requests or instructions, your Agent will search the Internet for information relevant to your needs, sort through it, then organize the findings and present them with corresponding references.

💡 Best practice

Remember to always fact-check the results you get from your Agent: AI systems can still hallucinate some of the answers that might appear plausible at first sight. We make it easier for you to check the results you get, each generated response comes with its corresponding sources you can simply click on.

Work With Multiple Data Sources

You can either select Internet to run an extensive search on the whole internet, or filter your preferences with certain sources only.

It is possible to switch between other sources such as News, Google Scholar, X, Reddit.

This makes it easier for you to find the information you need, no matter where it is located on the web.

Limitations

• The web search feature requires URLs to be publicly accessible.
• The content you fetch and utilize from a specific web page is not updated in real-time after the content is updated on the web page itself: if you're adding any information or URL to your knowledge base, we always suggest to check for online updates to make sure your information is up to date as well. 
• When you prompt your Agent to collect information from a web URL, you will only get the information contained within that page: if you input the link to a homepage, the Agent will not automatically scrape and collect the information contained in the entire website. 

💡 Pro Tip!

You can export web pages as PDFs and upload them as sources within your knowledge base.

Templates: Save your prompts and reuse them with ease

With templates, you can save your prompts and set custom input fields.

This means that you don't have to meddle with copy/pasting long prompts from your Notes app and changing certain words in parentheses each time.

The template above only shows 3 input fields: Topic, Keywords, Writing Tone. However, there is a hidden prompt in the back even though it's not visible:

This way, you can work with the same prompt over and over again for different cases. We allow you to customize your prompt however you like.

You can always go back to edit your template, too: adjust the instructions of the prompt behind your template and change the input fields whenever you need to!

Create your first template

1. Click on the "Templates" section of the TextCortex sidebar and then on "+ Create".

2. From there, you'll need to fill out some basic fields, such as the name and description of your template, but the most important aspect of your template will be the "Template Prompt".

A Template Prompt is the pre-built recipe for your template to work as you wish it to. With your instructions, you'll define what the AI should do, how it should do it, and provide some dynamic input fields that can be filled out later.

📌 Template Example

For example, let's say you want to create a "Caption Generator" for your social media posts. Your template might look like this:

• Template Name: Caption Generator
• Template Description: This template is used to create social media captions for posts.
• Template Prompt: Act as a caption generator for social media posts. Create a social media caption for [Topic Name] and make sure to write the caption in a [Tone] way for [Target Audience].

The inputs in square brackets are dynamic input fields that can be filled in every time you start a new interaction with this specific template.

As you start configuring your template, you'll see that the "Preview" modal adapts in real-time, allowing you to test how it performs based on your prompt and provided inputs.

Image Upload & Analysis

It is possible to upload images right within our platform and then prompt your Agent to analyze them.

To upload an image, you can either easily drag and drop it within the chat interface or click on the paperclip button and then on "Upload from computer".

Imagine uploading a picture of your product and asking your Agent to come up with an effective product description for your website or social media without having to describe the product itself in detail first.

You can also upload annual report visuals or any kind of table image embedded in a PDF document: your Agent will automatically take them into account once prompted.

Text-to-speech & speech-to-text capabilities

TextCortex offers text-to-speech capabilities on generated responses. On our platform, you can simply click on "+ Action" and hit Listen: the text will be read out loud.

If you'd like to make use of the speech-to-text funcitonality, you can simply click on the microphone button within the main chat interface, allow the necessary permissions inside your browser and start speaking.

Audio Transcription

The audio transcription feature makes it easier than ever to integrate audio content into your workflow: you can easily upload your files within our platform and obtain a transcription within minutes.

Supported Formats

• ✅ MP3
• ✅ MP4
• ✅ M4A
• ✅ Additional common audio/video formats

Get Started

To upload your file, click on the paperclip button and select "Transcribe Audio/Video".

You will then be able to upload your file and not only will it be transcribed swiftly: you’ll also get a transcription preview while you wait.

Once transcribed, you can upload the text into your knowledge base and let your Agent fetch information from your content whenever needed.