AI Employees

An AI Employee is a named coworker with a role, a voice and a set of skills. You talk to it in a chat. Quick questions get an answer straight away. Real work becomes a task: the employee writes a plan, waits for your approval when its settings require it, runs the plan in the background using the agents, tables, knowledge bases and tools you gave it, and reports back with the result.

Hire an employee when you want someone to own a stream of work over time: a sales development rep that builds account lists every morning, an account manager that answers pipeline questions, an incident coordinator that keeps your records in sync. For a single fixed procedure, aSequential Agent is simpler. For a chat that routes questions across specialists, use an Agentic Team.

How a message becomes work

Every message is read first. A question is answered in the chat. A request for work creates a task with a plan. Depending on the employee's Autonomy Scope, the plan either runs at once or waits for you to approve it. While it runs, any tool call that a compliance policy holds pauses the task until someone approves it.

Hire and set up an employee step by step

AI Employees live under Build, AI Employees. The editor is one long page of numbered sections. Section 1 must be saved before you can open the others.

1. Open the AI Employees list

   The list shows everyone hired in the workspace. Filter with All, On Duty (working on something now) and Idle, search by name, role or department, sort, and switch between grid and list. Each card shows the employee's role and personality, how many agents, tables, knowledge bases and tools it has, its tenure, and its Completed,Active and Failed task counts. A new employee shows an onboarding checklist instead of recent activity. Configure opens the editor andAssign work opens the chat.

   app.turtleaicoworker.com/employees

   

2. Import a ready-made employee (optional)

   Import from Library opens pre-built employees from official packs, partners and your organization. Filter by source or tag and search. A card whose model is not available in your workspace shows Choose model so you can pick one of yours; otherwise it showsImport. Open a card to read its role, personality, welcome message and instructions, and to see every agent, table and knowledge base it brings. Click Import N items, then View AI employee or Done.

   

   

3. Section 1, Identity

   Click Hire AI Employee. Enter a Name (required, up to 60 characters), a Role / title, and a One-line description (up to 200 characters, shown in the list and in search). Click Save employee. The other sections open once Identity is saved. The footer shows whether you have unsaved changes.

   app.turtleaicoworker.com/employees/new

   

4. Section 2, Welcome & voice

   Pick a Personality preset: Professional, Friendly,Casual, Formal or Custom. Refine it inPersonality & communication style, a free-text description that keeps the voice consistent. Write the Welcome message, the first thing people see when they open the chat; it supports markdown and bullet lists, and the Live preview shows how it will look.

   app.turtleaicoworker.com/employees/edit

   

5. Section 3, Skills

   Skills are what the employee can use. Each group has an add button that opens a searchable picker where you tick items and click Add Selected. The employee picks the right resource for each request on its own, so add fewer, well-described skills rather than many overlapping ones.

   • Agents (Add Agent): Sequential Agents the employee can hand work to. The Configure icon on a row lets you say when this agent should be used.
   • Knowledge Bases (Add KB): reference material it can search. Configure sets when to use it and how many results to pull.
   • Tables (Add Table): each row shows Read, Create, Update and Delete badges. Configure turns each permission on or off and adds a note on how the table should be used. New tables start with Read only.
   • Tools (Add Tool): connected integrations such as email, CRM or search.
   • Custom Tool Collections (Add Collection): API tool collections built in the Tools section.
   • Custom MCP Servers (Add Server): your organization's own MCP servers. At most 2 can be active on one employee; remove or switch one off to add another.

   app.turtleaicoworker.com/employees/edit

   

   app.turtleaicoworker.com/employees/edit

   

6. Section 4, Instructions

   Custom instructions are optional. The employee already chooses agents and knowledge bases from what people ask. Use this field to enforce specific rules ("always cite the policy section") or workflows ("when a notice arrives, run Triage first"). You can point at a resource by name with an @mention, for example @[Company Policies KB] or @[Email Agent]. Use Expand for a larger editor.

   app.turtleaicoworker.com/employees/edit

   

7. Section 5, Where they work

   The AI Employee Status switch turns the employee on or off; when it is inactive it does not respond to any message. Under Channels:

   • Slack: connect a Slack workspace so people can mention the employee in a channel or run it with a slash command.
   • Private Link Sharing: a link anyone can use to chat with the employee after entering a password. Set a Password, optionally Max sessions, the Max messages / session (100 by default) and an Expires date, then click Create Private Link. Setting a new password later revokes every active session.

   The Danger zone holds Duplicate, Export Config andDelete.

   app.turtleaicoworker.com/employees/edit

   

8. Section 6, Triggers

   Triggers let the employee start work on its own. Recurring triggers lists each one with its schedule, timezone and next run, plus a status badge, pause, Run Now, run history, edit and delete. Click New trigger and pick a type:

   • Schedule: time based. Give it a Trigger name and Instructions (what to do when it fires), then choose Recurring with a Cron expression and Timezone, or One-time with a date and time.
   • Integration event: fires on an event from a connected tool. Pick the integration and describe what to do.
   • Generic webhook: for Zapier, Make, n8n or your own code. A signed Webhook URL is generated when you save. POST any JSON to it; the body is available to the employee as {{trigger.payload}}.

   Event and webhook triggers share two Guardrails: Debounce (collapse repeated fires within N seconds, 0 for none) and Rate cap (runs per hour, 60 by default; firings beyond it are skipped and logged). Click Create trigger.

   

   

   After you create a schedule trigger, a review panel opens with the plan the employee will follow on each run. Click Approve & Activate to switch it on, Modify Planto ask for changes, or Generate New Plan to start again. A trigger stays inactive until its plan is approved.

   app.turtleaicoworker.com/employees/edit

   

   The history icon opens Run history for a trigger: every run with its status, duration and any error, filters for Completed, Failed andRunning, and Export CSV.

   

9. Advanced: model, media and autonomy

   Open Advanced at the bottom of the editor. Under LLM Configuration, choose the Provider and Model the employee thinks with, and aTemperature between Focused and Creative. An employee with no model cannot run tasks.

   Image Generation and Video Generation are optional. Pick a provider and a media model and click Save to let the employee create images or videos;Disconnect removes it. These are separate from the chat model.

   Under Execution Settings, the Autonomy Scope decides what runs on its own and what waits for you:

   • Scope 1, Human in the Loop: every task waits for you before it starts, even read-only research.
   • Scope 2, Supervised Autonomy (the usual choice): reading tables, searching knowledge bases, web research and read-only agents run on their own. Creating or updating rows, and any tool call that writes or sends, waits for you.
   • Scope 3, Bounded Autonomy: free to read and write your own tables, knowledge bases and agents. Any connected app (Gmail, Slack, HubSpot and the rest) waits for you, even to read.
   • Scope 4, Full Autonomy: everything runs without asking, including sending mail and writing to connected apps.

   Write limit per task (50 by default) stops a task and asks once it has made that many successful writes, such as rows created or changed or messages sent. Reads never count; set 0 to remove the limit. Show planning in chat displays the plan and step ticks live, andMax actions per turn caps tool calls per message (0 for unlimited).

   app.turtleaicoworker.com/employees/edit

   

   Further down, Configuration history keeps a snapshot each time a task first runs on a new setup, with Restore to go back. Share with Users & Groupsgives chosen users or groups Run Only access: they can talk to the employee but cannot change its settings. Click Update employee to save.

   app.turtleaicoworker.com/employees/edit

   

Work with an employee in chat

1. Send a message or assign a task

   Open the employee from the list. Type in the box and press Enter. Ask a question and it answers in the chat. Ask for work, and it creates a task card in the conversation with the task title and status. If your request is unclear, it asks a question before it starts; answer in the chat and it carries on. Mention a time ("tomorrow at 6", "later today at 2:40 pm", "every Monday at 9") and it can set up a scheduled run instead, which stays off until you confirm it.

   The header has Export (download the conversation), Conversations(past chats, with Me and Admin views, search and a new-chat button),Memory, Activity (the statistics page) andConfigure.

   app.turtleaicoworker.com/employees/chat

   

2. Approve the plan

   When the autonomy scope requires it, the employee shows the plan as a card: the steps, the resources each step uses, and the estimated time and step count. Click Approve & run to start, Modify to describe changes and click Submit Changes for a revised plan, or Cancel. With Show planning in chat on, you see each step tick off as it runs.

3. Approve held tool calls

   While a task runs, a tool call that needs a person (because of the autonomy scope, the write limit, or a compliance policy) pauses the task and shows an Approval needed card in the chat. Click Approve to run the call or Deny to skip it. Review & edit opens the exact call so you can correct it first, for example a wrong recipient or a typo in a subject, and the corrected version is what runs. Open in Compliance takes you to the full approvals queue.

4. Follow tasks in the queue

   The Tasks panel lists every task in the conversation, filtered byAll, Active, Completed and Failed. Expand a task to see its steps, its summary and Open Artifact; the row icons copy the summary, cancel, retry or delete the task. The Audit Trail tab lists the tool calls behind the work. Opening a result shows when it was requested, how long it ran, tokens and cost, withCopy result and Download PDF.

   

5. Attach files and get files back

   Click Attach to add up to 5 files to a message: images (PNG, JPG, GIF, WEBP), PDF, and text files (TXT, MD, CSV, JSON, YAML, HTML, XML). The employee reads files up to 5 MB directly; larger files, up to 100 MB, are stored so a tool can send them on. When the employee produces a deliverable, it is saved as an artifact you can revise ("make it shorter") and download as a PDF, CSV or Excel file. With an image or video model attached, it can also create media. Files made earlier in the same conversation can be reused, so "send the image above on Telegram" sends that image instead of making a new one.

6. Check what it remembers

   Click Memory to open What I remember about you. Memories are kept per person and grouped as Conversation History, Preferences,Instructions and Facts. Standing instructions you give in chat ("always check the CRM before quoting a price"), your preferences and facts carry across all your conversations; conversation history stays with the conversation it came from. Your preferences take priority over the admin instructions when the two conflict.

   

7. Review performance and daily summaries

   Activity in the chat header opens the statistics page for a date range, viewed asMe or Admin: completion rate, tasks completed, cost per completed task, how often a person had to step in, time per task, tasks over time, a task funnel, tasks by source (chat, scheduled, tool trigger, private link, Slack), engagement and cost, and memory growth.

   app.turtleaicoworker.com/employees/stats

   

   Each night Turtle AI Coworker writes a daily summary for every employee: tasks completed and failed, messages handled, tokens and cost, and highlights. On days with activity, the person who created the employee receives it as the Daily Digest email, which they can switch off in their notification settings.

Task statuses

Reference: employee settings

Trigger fields

Related

• Sequential Agents: the agents an employee hands work to.
• Agentic Teams: several specialists behind one chat.
• Tables and Knowledge Bases: the data an employee reads and writes.
• Tools & Integrations and Custom Tools & MCP servers: what an employee can call.
• Models: connect the chat, image and video models an employee uses.
• Approvals and Policies, budgets & guard rails: holding work for a person.