1. Where does Atom get its answers? Does it draw information from ChatGPT’s knowledge pool?
Atom is a blend of Microsoft’s Azure OpenAI service and Atomicwork’s own proprietary AI capabilities. We use OpenAI to help Atom be a conversational star, but we do not feed Atom any data that you do not provide. Atom is trained only on the content you upload and does not draw from ChatGPT’s knowledge.
2. How does Atom know what the correct answer is? What if there are two answers?
Atom summarizes answers from multiple sources. Atomicwork identifies which document influences the answer more and attributes the answer to it.
3. What are document formats that Atom can learn from?
Atom can “read” the documents you upload and the links you provide.
Atom can read the following file types: PPTX, PPT, XLS, XLSX, CSV, TSV, DOC, DOCX, ODT, PDF, RTF, TXT, MD, HTML, HTM, XML, JSON, YML, and YAML.
Atom can also learn from your Notion pages or SharePoint documents if you connect your Notion or SharePoint account to Atomicwork. Atom syncs with SharePoint documents every 6 hours to update its knowledge and every 24 hours with Notion documents.
Atom cannot read images, URLs, or code in your documents/Notion pages. This means that Atom will not be able to parse text inside images, use URLs in messages, or follow URLs to learn from the content of those sub-pages or “learn” from code.
4. How can I make Atom “forget” something? I uploaded the wrong document.
We index documents in real time so if you delete a document from a folder, Atom will stop serving up answers from the document.
5. Can I design a dialog flow for Atom? Can I customize and prioritize intents or assign variables, fallbacks, etc.?
You cannot currently design a dialog path for Atom. Please let us know what you have in mind and we can take it into account for when we do work on it.
6. My document is hosted on Notion/Confluence/Coda/another-document-tool, and I don’t want to give the document a public URL. What do I do?
We take security very seriously and are working very hard on building integrations with knowledge management platforms like Notion, Confluence, Coda, OneDrive, and Sharepoint so that you can easily sync your documents with Atomicwork. Until then, unfortunately, you will need to either download your document as a PDF and upload it or publish your document and give us the public URL.
7. What do you use to store the metadata and actual data/document upload?
We’re implementing robust security measures to protect user data and ensure the utmost confidentiality. Our data is safeguarded through encryption at rest and is contained within a closed VPC (Virtual Private Cloud) with no external internet access. We access this environment exclusively through a secure bastion host.
8. Who has access to my data and account?
When an Atomicwork account is created for you, you can create private workspaces for each team so that only the workspace users can access that team’s requests, documents, form answers and service catalog. Every workspace can have agents and admins to manage the workspace. Here’s how you can manage roles in a workspace.
9. Do you have access to my data?
Access to our database systems is strictly controlled and limited only to our CTO and head of AI. We also have stringent procedures in place for data access - we will access only on your request and for support purposes.
10. Do you use my data for training purposes? Is my data stored securely and separately? How can I ensure that my data remains within my control and complies with data residency regulations?
Atom has been built using the help of OpenAI and Atomicwork’s own proprietary AI capabilities so that it can hold its own in any conversation. However, your data will not be used by anyone except your team. We do not pass on your data back to OpenAI for training purposes.
Atomicwork follows a strict multi-tenant model. Our data integration is seamlessly orchestrated through the utilization of Weaviate DB, a self-hosted vector database, thoughtfully partitioned to ensure the utmost security and isolation for each valued customer. Furthermore, we actively pursue SOC2 and ISO compliance to fortify security across all dimensions.
11. Can I use my model and connect it to Atomicwork?
While we have this feature on our roadmap, we do not currently support BYOM (Bring your own model). Please write us your requirements, and we’ll do our best to help.
We also do not help our customers build their own models to plug into Atomicwork.