Enterpret Agent turns a question about your customers into finished work. Ask it what's driving churn this quarter and it researches across your feedback, reasons through the patterns, and comes back with a cited answer, or builds the report, ticket, or deck you asked for. You work with it live in a Session, or hand it an Agent Automation that runs on its own and delivers to Slack or email on a schedule.

If you knew this product as Wisdom, Enterpret Agent is what it became. The Context Graph underneath and the citation on every claim carry over. What's new is the working layer on top: it runs reusable Skills, connects to the tools where your work already happens, and produces real files, not chat text alone.

This article covers what Enterpret Agent is, how to find it, how to start your first Session, how to read the trail of work it shows you, and where to go next.

Enterpret Agent runs in two modes. Both are the same agent. The difference is who is steering.

Sessions (foreground). You drive the agent turn by turn. Ask a question, invoke a Skill, watch it research, steer it toward what you need. A Session is where you do interactive analysis and one-off work.

Agent Automations (background). You hand the agent a standing job and it runs on its own. An automation takes an objective, runs it on a schedule or in response to a signal, and delivers the result where your team works, such as a Slack channel or an email. The recurring report that you used to log in to build now arrives on its own.

You reach each mode from a separate entry in the left navigation.

Example questions to start with:

Two controls sit in the command box before you type:

A Session holds its context, so you can dig deeper without restating what you already asked:

Enterpret Agent shows its work as it goes, turn by turn. The trail is a live record you can follow and verify while the answer builds, not a finished block of text that lands all at once.

Thinking and steps. As the agent works, it lays out what it is doing turn by turn: the question it is answering, the steps it is taking, and the progress it is making.

Tool calls. When the agent reads your feedback, queries the Context Graph, or acts in a connected tool, that call appears in the trail. You see what it touched and what came back.

Sub-agents. For a large question, the agent delegates research to one or more specialist sub-agents that work in parallel and report back, then folds their results into a single answer. The trail shows this work happening.

Citations. Every count and every quote carries a link back to the source. Click an aggregate number to open the underlying records in your Enterpret dashboard, filtered to exactly what drove that number. Quotes are verbatim, pulled from the source with speaker attribution, never paraphrased. When there are zero results, the agent reports that plainly. It will not invent a number, a quote, or a link.

Because Enterpret Agent can do more than answer, it checks in before doing work in an external system. By default, it asks first ("Can I send this Slack message?") before taking an action through a connector. You can also switch a Session to run autonomously, without pausing on each action. Either way, the work trail keeps a record of what it did.

A Session does not stop at chat text. Ask the agent to produce a file and it writes one:

Citations carry through into the files, so a PDF is as verifiable as the answer it came from. You can also upload your own flat files (CSV, text, and data files) into a Session and have the agent analyze them alongside the Context Graph. Each Session holds a limited number of live files at once, and each file has a size cap.

Sessions are saved automatically and private to you. Share one with teammates and they can view your analysis and continue it with their own follow-ups.

Three surfaces give Enterpret Agent its working layer. Each has its own guide.

Skills. A Skill is a named, reusable workflow: your best analysis, encoded once and run the same way every time. Invoke one by typing a forward slash in the command box, for example /deep-dive-analysis on why mobile churn is rising. You can also ask in plain language and let the agent pick the right Skill, and you can create your own.

Most organizations get platform Skills out of the box for the core analytical jobs.

Connectors. Connectors let the agent read from and act in the tools where your work already happens. Each user connects their own accounts, because the agent acts with your authority. A Slack message it sends on your behalf is attributable to you and scoped to your access.

Agent Automations. When a job recurs on a cadence and the agent has the context to act, hand it off as a background automation. See the Agent Automations guide for setting one up.

You can also query your Context Graph from outside Enterpret. The Enterpret MCP Server brings your feedback into Claude, Cursor, Notion, and other AI tools. The MCP server serves the Context Graph; Skills and Connectors live in Enterpret Agent.

What happened to Wisdom? Enterpret Agent replaced it. Everything good about Wisdom carries over: the Context Graph, citations on every claim, and Rules that steer the agent's behavior. The new layer on top adds Skills, Connectors, a visible work trail, and real file outputs.

Which model does the agent use? The model picker lets you trade speed for depth: a faster model for quick asks, a more capable model for heavy comparative work. The choice locks in when you send the first message of a Session.

Does the agent ever make up a number or a quote? No. Every claim is grounded in your Context Graph and linked to source records. Quotes are verbatim. When a query returns nothing, the agent says so rather than spinning it as proof a problem does not exist.

Questions or feedback about Enterpret Agent? Reach out to your Enterpret team.