AI enterprise search in Athena answers one plain-language question across email, Jira, Confluence, meetings, SharePoint and web apps that have no API. Athena Intelligence’s agents search each app with the person’s own access, return an answer with a citation chip for each source, and hold any reply for approval before it goes out.
Most teams already have a search box in every app. The time goes to opening five of them, working out which one is current and writing it up for a waiting customer. Athena takes that work on from one chat.
Athena Intelligence, founded in 2022, builds an agent platform for regulated enterprises that can be self-hosted, where people and agents build agents, workflows and applications together. This dispatch follows one question about a customer’s single sign-on (SSO) rollout from start to finish. Figure 1 shows the whole idea on one page.
- One question in chat searches email, Jira, Confluence, meetings and SharePoint, each with the asking person’s own access.
- Web apps with no API are covered too. The person signs in once and saves a browse profile, and Athena does the rest.
- Answers carry citation chips, and hovering a chip shows the source line.
- A drafted reply waits for approval and is sent from the person’s own inbox. Every name in the video is demo data.
Contents
01What problem does AI enterprise search solve?
AI enterprise search removes the tab-by-tab hunt that comes before an answer. The facts usually exist, but they sit in five or six systems, and someone has to open each one, compare them and write the reply.
The number of systems keeps growing. In Gartner’s 2022 survey of 4,861 digital workers, 47% said they struggle to find the information they need to do their jobs well, and the average worker used 11 applications, up from six in 2019. A 2012 McKinsey Global Institute report estimated that interaction workers spent 19% of their workweek searching for and gathering information. Email and files are where much of it starts: in 2025, 85.15% of EU enterprises that buy cloud services used cloud email, and 71.53% used cloud file storage.
In the demo, an account lead named Dana needs to know where a customer’s SSO request stands before replying to the customer’s contact, Jordan. The pieces are spread across an email thread, a Jira ticket, a Confluence guide, last week’s meetings, SharePoint files and the customer’s supplier portal, which has no API. Figure 2 sets the manual version beside the Athena version.
02Which apps can AI enterprise search in Athena reach?
Athena searches the apps a team connects, any MCP server the team adds, and web apps with no API through a saved browse profile. Each person connects with their own account, so every search runs with that person’s access. Figure 3 shows the three ways in.
In chat, Athena searches email, Jira, Confluence, meetings and SharePoint. The integrations page also connects calendars, OneDrive and Google Drive, and Athena offers more than 150 data connectors. The connectors page lists them. A team can also attach a Model Context Protocol (MCP) server from the chat. MCP is an open standard for connecting AI tools to other systems.
Some systems will never get an integration, such as a customer’s supplier portal. For those, the person signs in once in Athena’s browser and saves the session as a browse profile. When a later question needs that site, Athena finds the saved login, opens the browser with the profile and reads the page. Athena has run browser agents in production for one to one and a half years, and one browser run lasted 25 hours and updated 70,000 CRM records. The research and browser page covers the browser in more depth.
Legal teams searching a document management system have their own walkthrough in Dispatch №13, AI search for your DMS, and the enterprise knowledge search use case covers the wider pattern.
03How does AI enterprise search work in Athena? One example, start to finish
The video runs just over a minute, with music and no voiceover. The first part connects the apps, and the rest follows one question from chat to a sent reply. Figure 4 breaks that question into six steps, with a checkpoint where a person approves.
- Connect the apps your team uses. Eight apps show as connected, including SharePoint, Confluence and Jira.
- Plus any MCP server you add, through Add Custom Server.
- Apps with no API: sign in once. Dana signs in to the Kestrel supplier portal inside a browse profile and clicks Save Profile, and the profile reads Authenticated.
- Ask across anything. Dana asks where Kestrel’s SSO request stands, naming email, Jira, Confluence, last week’s meetings and the supplier portal, and asks for a draft reply to Jordan.
- Agents search with your permissions. Five searches run: 4 emails found, Search Jira Issues, Search Confluence, Found 2 meetings and Found 3 SharePoint files. The portal returns Login required, so Athena opens the browser with the saved Kestrel Portal profile and reads the security review page.
- Every answer cited. The answer cites the email thread, the portal’s security review, the SSO rollout guide and the Kestrel weekly sync meeting. Hovering the Re: SSO timeline chip shows Jordan’s line, “We need SSO live before our Oct 15 rollout.”
- Nothing sends until you approve. An Approval required card shows the recipient, subject and message, and Dana clicks Approve.
- Sent from your own inbox. Jordan receives Dana’s reply and answers, “Perfect, thanks. Friday works for us.”
Table 1 compares the manual way with the Athena way, task by task.
| Task | The manual way | With Athena |
|---|---|---|
| Find the latest email | Search the inbox and scroll the thread | Email is searched with your access, and the thread is cited |
| Check the ticket | Open Jira and filter by customer | The same question searches Jira |
| Find the setup doc | Hunt through Confluence | The rollout guide is cited in the answer |
| Recall what was promised | Rewatch the call or ask a colleague | Meetings are searched, and the call is cited |
| Read a portal with no API | Log in, click through, copy the line | Sign in once and save the profile; Athena reads the page |
| Reply to the customer | Write from scratch and paste links | A drafted reply waits for approval and goes out from your inbox |
04How do you check an answer before acting on it?
Hover the citation. Each chip in the answer points at its source, and the hover card shows the excerpt with the quoted line highlighted. Figure 5 shows the four views a person sees in the product.
Answers are only as safe as the access behind them. Searches run on each person’s own connection, Athena does not store data from connected sources, and revoking someone’s access takes effect in real time. Human approval can be required on any tool, which is why the reply in the demo waits for a click. Most companies want that control: in a 2025 Gartner survey of 360 IT application leaders, only 15% were considering, piloting or deploying fully autonomous AI agents. The permissions and identity page goes into detail.
05Is Athena secure enough for regulated companies?
Yes. Athena Intelligence holds SOC 2 Type II and HIPAA, runs third-party penetration tests, has zero data retention agreements with Anthropic, OpenAI and Gemini, and does not train on customer data. Search and browse are also covered by zero data retention.
Athena deploys in a managed cloud, in your own VPC on AWS, Azure or GCP, on-premises, or air-gapped, including US government air-gapped networks and AWS GovCloud. A Big Four audit firm has run Athena in production for over a year, and Athena passed over a dozen enterprise security and AI vendor reviews this year. The deploy anywhere page covers the options.
06How do we get started?
Start with one question your team answers by hand every week, such as where a customer request stands. Connect the three or four apps where that answer lives, save a browse profile for the portal that has no API, and ask in chat.
Curious how it would work on your own apps? We’re happy to show you.
07Questions teams ask
What is AI enterprise search?
AI enterprise search finds information across a company’s connected systems, such as email, ticketing, wikis, meetings and file storage, and uses AI to assemble one answer with links to its sources. In Athena, the search runs from a chat with each person’s own access, and the answer carries a citation chip for each source.
Can Athena search web apps that have no API?
Yes. A person signs in to the web app once in Athena’s browser and saves the session as a browse profile. When a later question needs that site, Athena finds the saved login, opens the browser with the profile and reads the page. Athena has run browser agents in production for one to one and a half years.
Will people see information they are not allowed to access?
Searches run with each person’s own connection. Jira and Confluence, for example, connect with the person’s own Atlassian identity, so searches and reads follow that person’s permissions. Athena does not store data from connected sources, and revoking someone’s access takes effect in real time.
How do I check where an answer came from?
Every source behind the answer appears as a chip in the text. Hovering a chip shows the excerpt from the source with the quoted line highlighted, and an email chip opens the message itself. A reader can check each claim against the email, page or meeting before acting on it.
Does Athena send anything without approval?
Sending email can be set to require approval, and human approval can be required on any tool. In the demo, the drafted reply shows an Approval required card with the recipient, subject and message, and nothing sends until the person clicks Approve. The approved reply goes out from that person’s own inbox.
Where does Athena run?
Athena runs in a managed cloud, in your own VPC on AWS, Azure or GCP, on-premises, or air-gapped, including US government air-gapped networks and AWS GovCloud. Athena Intelligence holds SOC 2 Type II and HIPAA, has zero data retention agreements with Anthropic, OpenAI and Gemini, and does not train on customer data.
Run this on a schedule
Put the same questions on a schedule and get the answers as files.