Frequently Asked Questions (FAQ)

Simply click on + Create Deal

Give your deal a unique name

Deal code, Description, and Company ticker are optional fields if you want to add context to a deal to share with the rest of your team.

Select the privacy settings for the deal. You will have the option to choose between Firm-wide and Private

Hit Submit and your deal has been created.

Firm-wide: When a Deal or Chat is flagged as firm-wide, all users in the client instance (your company) can see the activity.

Private: Only the user who created the Deal/Chat can see the activity.

Firm-wide deals can have private Chats, which means the users all have access to the documents uploaded, but specific Chats are only visible to the creator of the Chat.

Firm-wide Chats, while visible to other users, can only be added to or edited by the person who created them.

Click the three-dot menu next to the deal name and choose Make Private. The current privacy status is displayed in the Deal box.

You can only edit deals that you have created. The three-dot menu will be grayed out for deals that other team members have created.

Click the three-dot menu next to the deal name and choose Control User Access (this is only available once a deal is private).

Then, select the users you want to include in the deal and navigate back to the deal page. These users will automatically have access once selected.

Yes, we have a “Favorite” setting available on the Deals page to make it easier to find the Deals you are involved in regularly.

To favorite a Deal, click on the star in the top right corner of the deal, to the left of the three-dot menu.

Then, once a Deal has been “Favorited,” you can navigate to the top of the Deals page and toggle between “All Deals,” “My Deals,” and “Favorites.”

On the deals page, you can sort deals by “All Deals,” “Active Deals,” “My Deals,” “Favorites,” and “Archived.”

Select the view you want to use at the top of the Deals page to change between these views.

All Deals: This display shows all the deals you can access within your environment. This includes Firm-wide deals, your private deals, and private deals shared with you. This also includes both active and archived deals.

Active Deals: This view only shows active deals and does not include archived deals. You will see all active Firm-wide and private deals that you have access to.

My Deals: This view shows deals that you have created. Deals that your team members created will not be visible in this view.

Favorites: This view shows deals that you have “Favorited.” Most users favorite deals that they are involved in regularly.

Archived: This view shows all archived deals.

If you are no longer working on a deal, you can archive the deal to remove it from the “Active Deals” view. Archived deals will still be displayed in the “All Deals” view.

You can only archive deals that you have created. To archive a deal, click the three-dot menu next to the deal name and choose Archive Deal.

After selecting Archive Deal, you will see the deal status (in the top left corner of the deal) switch from “Active” to “Archived.”

When you create a new deal, you will be auto-directed to the Upload Page. Here, you can drag and drop files into the upload box or select files using the standard Windows/Mac flow.

To upload documents to an existing deal, first select the Deal. Then, select Files from the menu. Finally, select Upload + in the top right-hand corner.

Either drag and drop files into the Upload box or select files using the standard Windows/Mac flow.

ALWAYS wait for the blue spinning circle to disappear and display “Complete” before paging away from the upload screen.

The circle disappears quickly with a small number of documents and can be present for up to a minute with a large list of documents.

All documents go through the following phases: pending, extracting, embedding, and ready. If a document doesn't make it to ready, it won't be used for any results. Sometimes, a document gets "stuck" (please tell us by reporting it through the Bug/Feedback option). Other times, if it fails, it will return a red "Error" code.

Pending - Initial load

Extracting - Retrieving content from a document

Embedding - Storing in Vector Database

Ready - Can be used for Q&A

Error - Failed to load

Yes, you can drag and drop folders into ToltIQ, and the files within the folders will be uploaded.

Yes, you can upload a .zip file using the same flow you’d use to upload an individual file or a folder.

From the Files page, select Download SEC Files.

Enter a ticker symbol and select Query.

A list of companies/entities will be listed.

Select one company or entity, and a list of 10K, 10Q, 8K and /A documents will be listed.

Select all or use the checkboxes to select individual documents.

Select Download Files.

The files will begin to be downloaded and will appear on the Files page, along with all other documents that have been uploaded (this can take up to 30 seconds, depending on how many were selected).

The SEC documents will go through the same upload process: Pending, Extracting, Embedding, and Ready.

ToltIQ defines and auto-assigns Categories using generative AI, which analyzes the document's content to determine the best fit.

Tags are also auto-assigned by ToltIQ based on the contents of the file uploaded. However, unlike Categories, the user can assign new tags, which can be anything that makes sense to organize the content better. You can create and delete as many Tags as you want.

Yes. Use the Document Select box on the left side of the screen to select all the documents you want to assign to a specific Tag. Then, select Batch Actions from the menu bar (which will indicate the total number of documents selected), and click Add Tag.

Type or Select the Tag you want to use and select Submit.

MS Word (DOCX)

MS Excel (XLSX)

MS PowerPoint (PPTX)

PDF

Web (HTML, HTM)

Image (PNG, JPG, TIFF)

Audio (FLAC, MPEG, MP3, MP4, MPEG, MPGA, M4A, OGG, WAV, WEBM)

CSV (Data Chat only)

ZIP

No, ToltIQ does not currently handle password-protected files.

If you would like to upload a password-protected file to the system, save a version of the file without password protection.

If it is impossible to remove the password protection, use Microsoft Print to create a new PDF version of the file and save it without a password.

On the files page, you will see a column named Extracted Tables. We will automatically identify and extract tables from any documents you upload and make them available for download.

If the document contains charts and/or graphs, we will also automatically extract the data and make that information available for download.

Any document with tables, charts, or graphs will have a Download button in the Extracted Tables column, which will download an Excel workbook with multiple tabs representing each table, chart, or graph ToltIQ identified.

Each table has a Raw/Clean pair.

On the Files page, you will see an option for File Explorer View on the top right corner of the page.

This view will display the folder structure based on how you uploaded the documents, where the hierarchy of the uploaded folders will remain intact.

“Traditional” data points for the files, such as file name, file date, and file size, are also displayed using the list view.

You should use Chat when you are interested in a “conversation” with context history with the documents you uploaded.

You want to get information from different documents, but you aren't sure where the content exists.

You want to have a specific chat with a "restricted" set of documents that you select based on categories, tags, or document names.

Follow-up questions consider the context of the conversation, including the responses to previous questions, which can lead to artificial "bias." If this seems to be happening, then start a new chat or turn Chat Context off.

A Restricted Chat limits the documents that will be used for Q&A.

There are three types of restrictions (Category, Tag, and Document list).

Once you select one or more Categories or Tags, only those documents labeled with that information will be included in the Chat. However, if new documents are added after the Chat has started, those new documents will automatically be included in future Q&A.

Selecting one or more documents restricts your Chat to those documents, and additional documents can not be added to the conversation.

You can view the restricted documents by clicking on the Select a document to view field at the top of the Chat.

Yes (see details on restricted Chat).

Yes, but only from that date/time going forward.

Yes. If you don’t Restrict a Chat, your Chat will contain all the documents you have uploaded for that Deal.

NOTE: Answers will often not be as accurate in a broad conversation with all documents vs. Restricted Chats. We recommend restricting your chat whenever possible.

First, pick a deal. Then, hit the + Create a new Chat button and name your chat. If you wish to limit the documents for Q&A, select the Restrict Chat documents option. Finally, click Confirm.

Yes. Click the three dots next to the chat name, select Edit, choose the visibility option (Private or Deal Team), and click Save.

Chats with the Deal Team visibility will be visible to everyone within your firm who has access to the deal.

Yes, use the toggle option at the top of the “Your Chats” menu to switch between “All Chats” and “My Chats.”

All Chats will display all of the chats within the Deal set to Deal Team visibility, including chats that you and your team members have created.

My chats will only display chats that you created within the deal.

First, select a deal and a chat. Then, type your question in the message box at the bottom of the screen, press Send, or hit Shift + Enter to send it.

First, select the gear icon to the left of the message box at the bottom of the screen. Then, select the LLM Model option above the message box and select a model from the drop-down list.

Select a Deal. Then, Select the box with the three dots to the right of the Chat name. Select the Export format (Word, Excel).

Each format will export the Chat Q&A (the Excel format will also include additional meta-data, Deal Name, Chat, etc.)

The document will appear in the Download Files list for your browser.

Select a Deal from the main menu.

Select Exports from the Deal Menu.

Under the Chat tab, check all the Chats you want to export.

Select the type of document you want to use (Word, Excel) and select Export.

The document will appear in your browser’s download list.

First, select the gear icon to the left of the message box at the bottom of the screen. Then, you can toggle the temperature between Grounded and Balanced.

Adjusting the temperature in an LLM is a way to control the randomness and creativity of the responses generated by the model. Here are several reasons to adjust the temperature setting:

Responses can only be regenerated for the last question answered in a Chat.

After a question has been answered, select the Regenerate option, which is the icon immediately to the right of the thumbs up/down.

Seeking Clarity or Accuracy: If the initial response is unclear, ambiguous, or seems inaccurate, regenerating it can provide a clearer or more accurate answer.

Exploring Different Perspectives: Generative AI can offer different viewpoints or ways of explaining a topic. Regenerating a response allows you to explore these diverse perspectives, which can be particularly useful in creative or analytical tasks.

Improving Relevance: Sometimes, the first response may not closely align with the specific focus or context of the query. Regenerating the response can yield more relevant information.

Resolving Ambiguities: If the initial response leaves room for multiple interpretations, a regenerated answer might resolve these ambiguities, offering a more straightforward explanation or conclusion.

Addressing Incomplete Information: If the first response seems incomplete or doesn't fully address all parts of the query, regenerating can help fill in the gaps.

Enhancing Detail and Depth: Sometimes, a more detailed or in-depth response is needed. Regenerating can help achieve this by providing additional information or a more thorough analysis.

The thumbs allow the user to give ToltIQ feedback on the quality of a response.

If a response is downvoted, you can also select a specific reason from a drop-down list or type in your own feedback.

Downvoted responses will impact the ongoing context of the conversion, though they will remain visible in the Q&A session.

That depends on the model selected, as different models allow us to maintain different history levels. At a minimum, we keep track of the last 10 Q&A responses for context.

Chat History includes or excludes the prior conversation as part of responding to the next question asked.

Conversations, by their nature, create bias in responses. This bias can be viewed as good, bad, or indifferent, depending on the nature of the next question. What the user wants included is always at their discretion. It can be re-enabled at any time after it is disabled.

As a general rule, if your next question significantly changes topics or you are looking to ask the same question again but with slightly different wording, temperature, or retrieval options, we recommend un-selecting Consider Chat History so the response is free of conversational bias.

Chat History is on by default. To turn it off, select the gear icon to the left of the message box at the bottom of the screen. Then, you will have the option to unselect “Consider chat history.”

ToltIQ allows you to either enable or disable the use of your documents to respond to a question.

If you un-select Consider Sources, then ToltIQ will remain aware of the conversation up to that point, but the response to your question will be sourced from the LLM ONLY. You can switch it on and off as needed, but be mindful that it will impact the conversation with potential responses that aren’t footnoted.

Consider Sources is on by default. To turn it off, select the gear icon to the left of the message box at the bottom of the screen. Then, you will have the option to unselect “Consider sources.”

Yes. Un-Select Consider Sources, and then the question you ask will only use the LLM for a response. This does not access the internet and only uses the training data of the specific model (in the case of GPT4-128K, that data is up to April 2023, but it is not ALL data on the internet up to that point, just the training data, which is proprietary)

Agent Mode is a different way to answer your questions. It allows you to ask a question about something current, and it will search the web for you. You can also provide a link (URL) to an article and ask questions about the article.

These features require the "Consider Web" option to be enabled. If disabled, the outcome will be similar to having Agent Mode turned off.

To use Agent Mode, turn it on by clicking the icon on the left side of the chat box under the gear icon. Then, select the gear icon and choose Consider Web (Agent Only).

When you’ve successfully enabled the “consider web” option, the globe on the Agent Mode icon turns green. This indicates that you are ready to use the feature.

This feature is still in beta, but we encourage you to experiment with it and provide us with feedback. We plan on launching more capabilities soon.

The bubble icon allows you to ask a question directly of the LLM Only.

This Q&A is “stateless,” meaning it will not be a part of the conversation or considered for purposes of context (it’s basically a way to query the training data of the LLM related to the diligence you are working on without impacting the conversation (i.e., it’s a side conversation).

Suggested Follow Up Questions are auto-generated by ToltIQ based on the response to the last question answered. Three follow-up questions are generated each time. If you are interested in asking one of the questions, select it, and it will automatically be populated in the question box.

Follow-up questions remain as part of the answer, so you can return to them later for further diligence activities.

Depending on the complexity of the document and how we retrieve information, footnotes will get embedded during the streaming response.

However, embedded footnote notes are not always guaranteed (we are continually improving the consistency with which they are used). The source documents will always be listed after the response.

Bulk Query allows you to select some or all of your documents and ask the same question of all the documents selected.

All documents are listed, and you select the documents to be included in the Bulk Query process by checking the “all” box (to the left of File Name) or by checking the specific documents to be included individually.

You can filter the documents before you check the boxes by using the “Search documents” bar (above File Name) to filter the document list.

After selecting the documents to be included, select Ask Question, located at the top of the page to the right of “Filter by document tags.”

Either type your question in the box provided or select Prompts to select a pre-loaded question.

Select the Model and Temperature settings from the drop-down lists and select Ask to initiate the process.

A spinning wheel will appear on the Ask Question button while the answers are being populated. Once complete, the spinning wheel will disappear, and you can view your answers or start a new Bulk Query.

If you ask over 200 questions, Bulk Query will go into “background mode,” and answers will be populated once fully generated instead of streaming live.

Bulk Query maintains a history of all your questions, allowing you to review prior results. You can access your previous questions by selecting History at the top of the Bulk Query screen.

From the History screen, you will have the option to reload answers or download question history.

Bulk Query should be used when you have a specific question about some or all the documents within a deal. It is not for conversational mode; rather, it is for asking the same question separately about each document selected.

Bulk Query is useful when trying to extract specific terms or data elements from a document.

Bulk Query is useful for quickly summarizing documents or extracting common data points from similar documents.

Yes, multiple questions from the Prompt Library can be selected at the same time using Bulk Query.

Select the prompts you would like to ask, and then hit Send.

When sending multiple prompts, answers will not be streamed like they are when only asking one question. Instead, questions and answers will appear in the Bulk Query History section when complete.

Yes. Use the checkbox to the left of the documents to select those you want to be included in the Bulk Query.

Bulk Query answers can be selected using the check box next to each document/answer.

After selecting the documents/responses, select Convert to Chat.

You will be redirected to the Chats section, where a new Chat topic will be created based on your selection.

This new Chat will be a Restricted Chat based on the documents selected

You can now continue the conversation.

Select a Deal. Select Bulk Query and then select View History.

From the View History page, select Download Question History. This will bring you to the Exports page.

Alternatively, you can select the Exports page directly from the top menu bar.

From the Exports page, select Bulk Query from the menu bar.

Then, select Question History from the right side of the grid that appears on the screen and select the questions that you want included. Then, select Export to generate an Excel file with your content.

The Exports page allows you to export multiple Chats or Bulk Query outputs simultaneously.

Document Summaries are created at the user’s request.

The list of all documents uploaded is available on the Document Summaries page. The document names are “enhanced” through generative AI to make the name more informative.

The user selects a document and then initiates the dynamic summary creation by selecting Generate New Summary.

After selecting Generate New Summary, the user will have multiple options:

Please note that the system can take several minutes to generate a summary. While the summary is being generated, you can leave the page to continue with other work.

If the user is not satisfied with the output, they can rerun the process by selecting Generate New Summary after the process has initially completed.

If you are satisfied with the summary but would like to add a section, use the Request New Point feature.

Navigate to the Document Summaries page within a deal.

Select Manage Summary Templates on the top right corner of the page.

To create a new summary template, select the blue Create New Summarizer Template button on the top right corner of the page. Then, input your template name, select the topics you would like to include in the template, and hit Save.

To edit a summary template, choose the pen icon in the Actions column. You will have the option to change the template name, edit topics, delete topics, and add topics. After making your updates, select the blue Update button to confirm your changes.

Data Chats use OpenAI’s code interpreter to get into a deep conversation with multiple spreadsheets.

Data Chats can be used to “talk” to the data in spreadsheets, similar to building formulas in Excel to get results.

Yes, when you create a new Data Chat, you will be able to select multiple spreadsheets.

Yes, to generate a chart, start by creating a Data Chat with the spreadsheet(s) you’d like to use as a source. Then, use the prompt “Create a [chart name] based on the data I uploaded.”

After the chart is produced, you can also clean and format data by prompting it to remove duplicates and missing values.

Area chart, bar chart, box plot, correlation chart, dendrogram, heat map, histogram, line plot, pie chart, radar chart, scatter plot, and venn diagram.

Select a deal. Select Data Chats and choose the chat for which you want to export table response data.

Any time a table is formulated as part of a response, you will see a blue download button in the top right corner of the table.

When you click the download button, the export will appear in the Downloaded Files list of your browser.

Select a Deal. Then, navigate to Data Chats in the top menu and select the box with the three dots to the right of the Data Chat name.

Select the Export format (Word, Excel, Markdown, JSON). Each format will export the Data Chat Q&A (the Excel format will also include additional meta-data, Deal Name, Chat, etc.)

The document will appear in the Download Files list for your browser.

Use the </> button to View Source Code. In a Data Chat chat, Open AI’s code interpreter uses the spreadsheet files as the source and dynamically writes Python code based on the question asked.

The </> button will allow you to view and analyze the dynamic Python code created by the system.

Data Chats are somewhat of a “black box,” so sometimes the dynamic Python code is not shown.

Select the Submit Bug button to the left of the ToltIQ logo in the top bar of every screen.

Yes (but it is optional). Select the User Icon in the top right corner of the screen, choose Manage account, and then select Security.

On the Security page, select Add two-step verification.

A manager is a regular user of ToltIQ.,

An admin has the ability to add and delete users.

Click on your company name on the upper right corner and click on the gear icon .

Select Members from the menu.

Select Invitations.

Select Invite.

Type in the email address of the user you want to add.

Select the role of the user (Member or Admin).

Select Send Invitations.

Click on your company name on the upper right corner and click on the gear icon .

Select Members from the menu.

Select the box with 3 dots to the right of a user.

Select Remove Member.

support@toltiq.com

Not directly once the Chat has been initiated.

If it is a Restricted Chat and you have selected Categories or Tags, then any new documents with those settings will get added from that point forward.

At the end of your question, say, “Use a table format.” If you want specific columns in the table, then list them as well.

At the end of your question, say, “Use bullet point formatting.”

We are dependent on the performance of the various models from OpenAI and Anthropic.

The Prompt Library is a tool that allows you to manage frequently asked questions across your diligence activities.

Yes, to sort prompts by category, use the Group Prompts feature.

Once Group Prompts has been selected, you will see that your prompts will be organized into “folders” based on the prompt category (10K, 10Q, CIM, etc.)

If you wish to view all prompts instead of grouped prompts, turn the Group Prompts feature off.

Select the Prompts menu item from the top left section of the main screen.

Select New Prompt.

Enter the Question you want to have in the library and assign it a Category (which will be used for sorting and filtering).

If you wish to initially populate your firm’s Prompt Library with standard prompts from ToltIQ, select Prompts from the top left main menu.

Select the Questions you wish to add to your unique instance of ToltIQ by checking all the ones you want to include.

Select Add Selected, and those questions will now be a part of your Client Prompts library.

You can then add additional questions using the New Prompt button.

You can delete questions using Remove Selected.

Create a CSV file with two column headers: text and category.

The first column will be the question, and the second column will be the category (any category you want).

Example: “How many employees did the company employ in 2022?”, “Corporate”

Select Upload in the top left corner of the Prompts page and upload the CSV you created with the list of questions you’d like to upload.

Once you have selected a Chat and are in the Question box, the Prompts option will appear above the Send button to the right of the chat box.

Select Prompts, and a list of your custom prompts will appear.

Select one or more prompts from the list (you can also filter the list)

If you select only one prompt, your options will be Populate or Send Message.

If you select more than one prompt, your options will be Populate, Send Serially, and Send Parallel.

Yes, select the moon or sun icon in the top right corner.

Yes, Select Cancel (located in the same place as the Send button) after a question has been asked, either before or while a response is being created.

Yes, our spot testing has shown impressive language translation from major languages; however, we have not rigorously tested this capability.

You can upload MPEG files, and they will be translated into a transcript that allows you to ask questions and get specific answers from the content.

If the documents are in a deal that has already been created, create a new Chat and enable the Restricted Chat setting to select the specific documents you want to include in the analysis. (If this is a brand new Deal, then create the deal first and upload the documents).

Once the Chat is established, determine the best model for the Q&A session. Refer to the table above for additional information on determining the best model for your specific needs.

Ask a question to compare the specific information you are interested in and specify the documents by explicit name (or if there is a unique aspect to them use that information instead of the full name ie: 2021 and 2022 ESG reports).

If you would like the response in table format, add the phrase “Use Table Format” at the end of the question before submitting it.

You can enable or disable the Sources feature at any time to have a more direct conversation with the LLM (e.g., GPT4). You will not get footnotes or sources from the answer, but if you set the Creativity/Temperature to 0, you should expect a reasonably accurate response based on the training data (for GPT4-128K, that would be December 2023).

Keep in mind that while the training data has included a wide spectrum of content in terms of breadth and depth, it is not real-time or connected to the Internet for ongoing updates.

When Sources are enabled, the answer will be grounded in the documents you have uploaded, but it will leverage the insights on the topics you are asking about to formulate responses. It also has a high degree of understanding of how questions are phrased to best respond.

The Retrieval Strategy is the embedding logic used to match your question to the contents in the documents and then retrieve chunks of data that will be refined to create an answer.

Generally, the new DotProduct (default) option will work great for most queries. However, as the number and complexity of documents grow in a deal and your questions become more complex, you may find the Hybrid Search is more effective if you aren’t getting good results with DotProduct:

Ada2 (Old Default): This is the traditional cosine model (often referred to as a similarity search) we've been using. It uses dense vectors to understand and compare the meanings of your question and the stored content. Dense vectors are numerical representations of the meaning behind words. It's reliable and good at finding semantically similar content, even if the exact words don't match.

Ada2 DotProduct (New Default): This is an improved version of the old method. It uses a technique called the "Dot Product" to quickly and efficiently compare your question with the content (for those that enjoy the technical details – the Dot Product calculates the sum of the products of the corresponding elements of the two vectors. It takes into account both the magnitude and the direction of the vectors). The dot product measures how similar two vectors are, which helps in retrieving the most relevant results for your question. This method can provide more accurate results by better understanding the nuances in your search queries.

Ada2 Hybrid Search (New): This is the most advanced option, combining the strengths of two different search methods. It uses both dense vectors and sparse vectors. Sparse vectors focus on keyword matching (like how many times specific words appear), while dense vectors focus on the overall meaning of the text. By combining these, hybrid search can get results that are both semantically relevant and keyword-specific. This method often provides the best of both worlds, offering relevant search results by considering both the meaning of your query and the specific keywords used. It's useful when you need both precise and comprehensive results.

“ABC” is the same as “XYZ” (this is in the rare case where the AI model may not realize the first term is similar to the second term, and you’re

Be brief

Be concise

Be detailed

Be specific

Be verbose

Don't elaborate

Enumerate

Extract the following information

Ignore xxx (where xxx is something you've seen in a prior response or is content you know exists in a document but isn't relevant)

Take it step by step

Take your time

Think carefully

Use bullet points

Use a numbered list

Use a table format

We recommend using Google Chrome or Microsoft Edge. We support Safari but would not recommend using it, as performance is demonstrably better using Chrome and Edge.

Yes, we support language translation from English, German, French, Spanish, Italian, and Portuguese for scanned documents, and for machine generated text, all languages can be read in chatGPT (70+ in total).