Core Actions
Autopilot does real work across the Ema platform on your behalf. Instead of clicking through menus and forms, you describe what you want in plain language and Autopilot takes the actions for you — building and editing AI Employees and their workflows, managing knowledge bases, connecting Tools, generating documents, and answering questions about how the platform works. This page is the reference for everything Autopilot can do.
An AI Employee (AIE) is a configured worker on the Ema platform; its behavior is defined by a workflow — a directed graph of typed nodes — plus its Instructions, Knowledge bases, and connected Tools. Autopilot can create and change all of these.
How Autopilot works. You drive Autopilot with a conversation. It takes one or more steps to fulfill a request — reading a workflow, adding a node, uploading a file — and streams its progress back as it goes. For anything destructive or hard to reverse, it pauses and asks for your approval first; see Permissions and scoping.
What Autopilot can do
| Capability | What it covers |
|---|---|
| Build and edit workflows | Open a workflow, add and remove nodes (agents), connect and disconnect them, rename them, set input values, and validate or deploy the result |
| Configure agents (AIEs) | Configure the agent inside a node, set per-agent Instructions, assign Tools, and choose an EmaFusion™ model |
| Create and manage AI Employees | Create, update, clone, and delete AI Employees, and organize them into groups |
| Manage knowledge bases | Upload data, list and search Knowledge bases, copy a data source to another AI Employee, and remove data |
| Connect Tools | Browse available integrations and actions, search them, and wire them into a workflow node |
| Work with dashboards | Upload dashboard data, view rows, and trigger a workflow run for a row |
| Generate documents | Produce a report or other deliverable and deliver it in the conversation |
| Test an AI Employee | Start a conversation with an AIE, send it messages, and review its replies |
| Ask platform questions | Get answers about Ema capabilities, features, and how things work |
| File a bug or request a feature | Send a bug report or feature request to Ema directly from the conversation |
The rest of this page covers each capability in detail, with example prompts you can adapt.
Build and edit workflows
A workflow is the directed graph of typed nodes that defines what an AI Employee does. Autopilot can build one from scratch or modify an existing one. Editing happens against a working copy — Autopilot loads the workflow, makes the changes you ask for, and nothing affects the live AI Employee until you ask it to deploy.
A typical edit session looks like this:
- Tell Autopilot which workflow to change and what you want.
- Autopilot loads the workflow and shows an overview of its current nodes.
- Autopilot makes the changes — adding or removing nodes, connecting or disconnecting them, renaming them, setting input values, configuring the trigger, or adjusting categorizer options.
- Ask Autopilot to check the workflow. It validates the configuration and reports any errors.
- Ask Autopilot to deploy. Because deploying pushes changes to your live environment, Autopilot asks you to confirm first.
As Autopilot works, the steps it takes appear inline in the conversation — for example "Opened workflow," "Added agent," "Connected agents," "Validated workflow." In expanded mode, a Workflow changes panel on the right summarizes what changed (nodes configured, edges wired, the deploy status), so you can track the edit at a glance.
Example prompts
Build me an internal knowledge assistant that answers employee questions about company policies and SOPs. Create a sample policy file, upload it, and test that it works. Make all the design decisions yourself.
Add a human-approval step to my "Invoice Processor" so someone on the finance team must sign off on payments over $10,000, then check and deploy it.
If you want Autopilot to handle a build end to end without stopping to ask, say so explicitly — for example, "make all design decisions yourself without asking me for input." Autopilot still pauses for approval on destructive actions and on deploy.
Configure agents
Each node in a workflow runs an agent. Autopilot can configure an agent's behavior in place: set its Instructions, assign the Tools it can use, and pick the model it runs on through EmaFusion™. You can also adjust a categorizer node's options when a workflow branches on a classification.
Example prompt
In my "Support Triage" workflow, give the classification step a higher-reasoning EmaFusion™ model and add the Zendesk Tool to the step that drafts replies.
Create and manage AI Employees
Autopilot can create a brand-new AI Employee, update an existing one, clone one, or delete one. It can also organize your AI Employees into groups, which act like folders — create a group, move AIEs into it, rename it, or delete it. Deleting a group leaves the AI Employees inside it ungrouped; it does not delete them.
When Autopilot creates an AI Employee, it reports back with a summary of the name, description, and configuration. Creating, updating, and cloning all require permission to create AI Employees in your tenant; see Permissions and scoping.
Example prompts
Clone my "Customer Support Bot" and rename the copy to "Support Bot v2."
Create a group called "Sales Team" and move my "Lead Qualifier" and "Demo Scheduler" AI Employees into it.
List all my AI Employees that are currently active.
Manage knowledge bases
An AI Employee answers from the Knowledge bases attached to it. Autopilot can add to and manage that data:
- Upload a file as a new data source. Attach it to the conversation, then tell Autopilot which AI Employee it belongs to. See Modes, feedback, and tips for the three ways to attach a file and the size limit.
- List the data sources attached to an AI Employee.
- Search a Knowledge base by keyword and summarize the top results.
- Copy a data source to another AI Employee.
- Delete a data source. Because this is permanent, Autopilot asks you to confirm first.
Example prompts
Upload the Q4 sales report to my "Revenue Analyst" AI Employee and search it for mentions of APAC.
Show me all the data sources attached to my "Policy Assistant."
Connect Tools
Tools let an AI Employee read from or take action in an external system. Ema connects to a large catalog of apps. Autopilot can browse what is available, search for a specific action, and wire a Tool into a workflow node when you edit an AIE.
- Ask Autopilot what integrations or actions are available.
- Pick one and ask Autopilot to add it to a workflow node.
- Autopilot configures the Tool on that node so the agent can call it at runtime.
Example prompts
What apps and Tools can I connect to Ema? Show me what's available.
Add a Salesforce Tool to the lead-lookup step of my "Lead Qualifier" so it can pull in new leads.
Credentials. Some integrations need their own credentials before an AI Employee can use them. Not every integration supports every action — ask Autopilot about a specific capability if you are unsure.
Work with dashboards
A dashboard is structured, row-based data you can run a workflow against. Autopilot can upload dashboard data, show you the rows, and trigger a workflow run for a specific row — useful for batch processing or for spot-checking how an AI Employee handles a particular record.
Example prompt
Upload this CSV as a dashboard for my "Invoice Processor" and run the workflow on the first row so I can see the result.
Generate documents
When you need a polished deliverable — a report, summary, or structured analysis — Autopilot can generate it and deliver it in the conversation. The finished file appears as an artifact you can preview and download; in expanded mode it also lands in the Generated list in the right rail.
Example prompts
Generate a summary report of all the support tickets handled by my "Support Triage" AI Employee this week.
Draft a one-page onboarding guide covering our PTO policy, expense process, and IT setup checklist. Use headers and bullet points.
Test an AI Employee
Building an AI Employee is only half the job. Autopilot can open a conversation with any AIE, send it messages as an end user would, and show you the replies — without your leaving the panel. You can start a conversation, send follow-ups to probe different scenarios, rename a test conversation, and delete one when you are done. Deleting a conversation is permanent, so Autopilot confirms first.
Example prompts
Start a conversation with my "IT Helpdesk" AI Employee and ask it: "How do I reset my VPN password?"
Create a 5-question test set for my internal knowledge assistant, run it, and report the results. If the AI Employee doesn't exist yet, create one first.
Ask platform questions
Autopilot doubles as a platform guide. Type a question directly — no special syntax — and it answers based on current platform capabilities.
Example prompts
Give me an overview of what I can do on the Ema platform and how you can help.
What's the difference between a scheduled trigger and a manual trigger for an AI Employee?
For billing, account limits, or enterprise-agreement questions, Autopilot may point you to Ema support rather than answer directly.
File a bug or request a feature
If something is not working as expected, or you want to suggest an improvement, you can send it to Ema directly from the conversation — no separate ticketing tool needed. Describe the issue or idea and Autopilot files it.
For a bug, include what you were doing, what you expected, and what happened instead. The more specific you are, the more actionable the report. Submissions go to Ema's internal tracking; you will not get a ticket number back in the chat, but every submission is reviewed.
Example prompts
I found a bug: when I upload a CSV to my "Data Analyst" AI Employee, the upload spinner never stops and the file never appears. I tried three times with different files.
Feature request: it would help to schedule a workflow to run on the last business day of each month, not just a fixed date.
What's next
- Permissions and scoping — when Autopilot asks for approval, and what it can access.
- Modes, feedback, and tips — Fast vs Thorough, the two views, attaching files, and giving feedback.
- Limits, availability, and troubleshooting — what Autopilot can't do and what to do when it's unavailable.