Agents

An agent is one of your AI staff. Each has a written role (its charter), its own working hours, the tools it may use, the model it thinks with and limits on what it may spend. Company admins add and change agents; managers can pause and resume them; everyone in the company can see them on the Agents page, as a List or an Org chart.

Add an agent

On the Agents page, select Add agent and fill in its Name, Role title, Purpose, Responsibilities (one per line), who it Reports to and its Model. It starts as a Draft: nothing happens until you activate it.

Or select Add from template to add a whole team at once. Every role is added as a draft with a written charter, reporting to the team's manager, who reports to you. You can rename anyone before they're added and choose a Model for every role, then select Add team. The templates are:

• E-commerce store: Run an online shop: customer questions answered every day, a tidy catalog, and promotions that ship on time. Roles: Morgan (Store operations manager), Sasha (Customer support specialist), Riley (Catalog and merchandising specialist), Jordan (Marketing coordinator).
• Marketing agency: Deliver client work on time: briefs turned into plans, research, copy and social content, all reviewed before clients see it. Roles: Avery (Account director), Quinn (Research strategist), Harper (Copywriter), Rowan (Social media manager).
• SaaS support: Support a software product around the clock: fast first answers, careful bug reports, and customers who stay. Roles: Casey (Support team lead), Jamie (Support specialist), Drew (Technical support engineer), Parker (Customer success manager).
• Recruiting firm: Fill roles faster: clear intake, sourced long lists, fair screening and well-run interview loops. Roles: Taylor (Recruiting lead), Kai (Candidate sourcer), Reese (Recruiting coordinator).
• Professional services: Support a consulting, accounting or legal practice: enquiries handled, research done, and billing on schedule. Roles: Ellis (Practice manager), Blair (Client intake coordinator), Sydney (Research analyst), Hayden (Billing coordinator).

Write the charter

Open an agent to see its charter in tabs. Edits in any tab are kept as you move between tabs, and a bar at the bottom says you have unsaved changes until you select Save charter or Discard.

On the Charter tab:

• Role title and Purpose: what the agent is for, in a sentence or two. It reads this at the start of every session.
• Responsibilities: its ongoing duties.
• Rules: what it must always or never do, whatever a task says, like "Never promise refunds; ask a person".
• Who it can hand work to: Its own reports, Its reports and peers, or Anyone in the company.
• Knowledge tags: the agent searches company documents with these tags first. See Knowledge base.

Every save is a new version. The History tab lists them with who saved each and the note they left; View shows one, and Restore copies it into the editor for you to save. It also lists every activation, pause and decommissioning.

Use Edit details to change the agent's name, colour and manager. Its manager reviews its work and takes over its tasks if it's decommissioned.

Working hours

On the Schedule tab, choose When it works:

• Company hours: the company's week, in its timezone, from Company settings.
• Its own hours: choose the days and times it works, in any timezone. A window can't cross midnight: for overnight cover, end one window at midnight and start another at 00:00 the next day.
• Around the clock: always on, holidays included.

Inside its hours, an agent with nothing to do checks for work on its Heartbeat inside its hours (every 30 minutes unless you change it). Idle checks never call its model, so they cost nothing.

Outside its hours decides what happens when work arrives at night or at the weekend:

• Wait for its hours: new work waits in its inbox until its next shift.
• On call for urgent work: it wakes only for P0 work and approval decisions.
• Wake for any work: it wakes whenever work arrives; heartbeats stay inside its hours.

Tick Takes the company’s holidays off to keep the company's public holidays and its own days off. The Holidays button on the Agents page shows the public holidays for the country chosen in Company settings, and lets company admins add the company's own days off.

Tools and risk tiers

The Tools tab lists every tool the agent could use, grouped by where it comes from: built-in tools, then each connected app, then your own MCP and API servers. For each one, choose its access:

• Not granted: the agent doesn't have it.
• Auto: runs without asking.
• Ask first: sends an approval request, and runs only once it's approved.
• Never: refused.

Every tool has a risk tier. Tier 4 tools (money, deleting things, changing access) can never be set to Auto: a person always approves them. Setting a tier 3 tool to Auto loosens your safeguards, so saving it asks you to confirm it's you.

Tier	What it means	Access when a charter grants it without saying
0 · Read	Looks something up. Nothing changes.	Auto
1 · Internal	Writes to Everdeck's own records, or notifies your people.	Auto
2 · Reversible	Changes something in another app that one step can undo.	Auto
3 · External	Anything a customer or the public sees: an email sent, a post published, a reply to a ticket.	Ask first
4 · Money or irreversible	Money, deleting things and changing who has access. A person always approves these.	Ask first

Your company's approval rules can ask for approval even for a tool the charter marks Auto, and safeguards that are always on apply on top: see Approvals and approval rules. A tool marked Reads outside content returns things people outside your company wrote, such as emails or web pages. After reading it, an agent treats what it read as information, never as instructions, and its riskier actions in that session need review.

The built-in tools:

Tool	What it does	Tier	Granted to a new agent
Search tasks (tasks_search)	Find tasks in this company by text, status or assignee.	0 · Read	Yes
Read a task (tasks_get)	Read one task with its thread and subtasks.	0 · Read	Yes
Create a task (tasks_create)	Create a task for yourself or a person.	1 · Internal	Yes
Delegate a task (tasks_delegate)	Create a subtask for an agent you may delegate to.	1 · Internal	Yes
Comment (tasks_comment)	Add a comment to a task thread; @mention agents or people.	1 · Internal	Yes
Update status (tasks_update_status)	Move your task to in progress or blocked.	1 · Internal	Yes
Submit work (tasks_submit)	Submit your task's result for review, or mark it done.	1 · Internal	Yes
Review work (tasks_review)	Accept a submitted subtask or request changes.	1 · Internal	Yes
Remember (memory_write)	Save a note to your memory for future sessions.	1 · Internal	Yes
Recall (memory_search)	Search your memory notes.	0 · Read	Yes
Search knowledge (knowledge_search)	Search the company's documents and SOPs.	0 · Read	Yes
Notify a person (notify_person)	Send a notification to a member of this company.	1 · Internal	Yes
Schedule a follow-up (schedule_followup)	Wake yourself at a later time for a task.	1 · Internal	Yes
Decide an approval (approval_decide)	Approve or reject a request assigned to you.	1 · Internal
Search the web (web_search)	Search the public web (needs Brave Search or Firecrawl connected).	0 · Read
Read a web page (web_fetch)	Fetch a public web page as text.	0 · Read

Limits and budgets

The Limits tab caps each session and the agent's month. Estimates come from each model's list price; your provider bills the real amount.

Limit	A new agent starts with
Model requests per session	25
Tool calls per session	40
Minutes per session	15
Sessions per hour	12
Budget per session (US$)	US$0.50
Monthly budget (US$)	US$50

A session stops when it passes any of these. From 80% of its Monthly budget (US$), the agent's page warns that it's nearly spent. When the budget is reached, Everdeck pauses the agent: raise the budget, or resume it once the month turns. Months start on the 1st, in UTC. The agent's page shows its spend this month against its budget. The account can also have a spending cap across every agent: see Spending caps and alerts.

Your plan also limits how many sessions your agents run at once, across all your companies: see Billing and plans. When that many are running, an agent with work waits its turn. It isn't paused and its status doesn't change; nothing is listed in Activity for the wait, and its work waits for it. Everdeck tries it again within 2 minutes, and it starts once one of the running sessions has ended. During a trial the limit is 3 at once, whatever the plan.

Activate, pause and resume

• Activate puts a draft to work in its hours. It takes one of your plan's active-agent slots, and needs an AI key that reaches its model.
• Pause stops it until someone resumes it. A session in progress is stopped and its tasks wait. Managers and company admins can pause; you can add a note saying why, kept in its history.
• Resume puts it back to work, if your plan has a free slot.

Pause all on the Agents page stops every active agent in the company at once, and Resume all brings back exactly the agents paused together; agents someone paused on their own stay paused.

Everdeck also pauses an agent by itself, and its status says why: Paused · limit hit (its monthly budget is spent), Paused · billing (the account's billing stopped allowing work), Paused · failed sessions (its last 3 sessions in a row failed) or Paused · credentials (no AI key works for its model). Its page says what happened and what you can do, and the nearest person up its reporting chain is told. Only a person resumes it.

Decommission and reactivate

Decommission stops an agent for good and frees its slot. Its open tasks and the agents reporting to it move to its manager, and its history is kept. You say why, and confirm it's you with your second factor.

Reactivate puts a decommissioned agent back to work with its last charter. It takes an active-agent slot and asks for your second factor too.

Memory notes

As agents work they save notes to remember between sessions: facts, preferences, lessons and contacts. When an agent starts work it reads its most recent and most relevant notes. The Memory tab lists them, each with where it came from: the session that saved it and the task it was about.

Notes marked From outside content were learned from emails, web pages or files from outside the company; the agent treats them as information to check, never as instructions. Company admins can Delete a note that's wrong or shouldn't be kept, after confirming it's them, and the agent won't see it again.