Coming next: model accounts, a Safety page and a shareable Task board
The screens for work whose groundwork is already in production: provider accounts and model choice, a trust ladder for what Agents may do alone, a read-only Task board link, workspace backups, clearer Agent status and a spending breakdown. None of this is in the app yet, and details and timing can change.
Each item below builds on platform work that is already in production but, for the most part, has no screen yet. This page lists the screens and controls that come next, as described in that merged work and its specifications.
Nothing here has a date. Plans can change, and an item may arrive in more than one release. When a screen reaches production, its entry moves to that release's notes.
On this page
Upcoming changes(6)
Several provider accounts, and a model for each Agent and schedule
A new Settings → Models screen will let you add more than one account per AI provider in the order you choose, set a default model, and pick a different model for one Agent or one schedule.
What is coming: a Models tab in Settings, a model panel in each Agent's settings, and a model choice for a single schedule.
- Provider accounts. Add several accounts for the same AI provider and put them in order. Rename, pause, resume or replace a key without losing the account's history. Agents use the first account in that order that still works.
- Account health. Accounts are checked in the background, and a banner tells you when one is about to expire or has stopped working.
- Which model does which job. Set a default model for the workspace, give one Agent its own, and let a schedule use a different model, so a quick hourly check does not run on the model meant for heavy work.
Already in place: provider accounts and these model choices exist in the platform today, and each Agent run records which model answered it, but for now they can only be managed through the API.
Limits: these choices apply to Agent runs. A key set on a Work stays in force, and generating a Work, the AI assistant and the Knowledge Base keep using the Work's plugin settings. Switching automatically to another account or model when a call fails is a later step.
A Safety page to decide what Agents may do on their own
Settings → Safety will show a trust ladder for each kind of work, such as emailing people outside, publishing or running commands, with Off, Draft, Ask and Auto rungs, and a log of what the platform stopped.
What is coming: a Safety tab in Settings and a Safety tab on each Agent.
- The trust ladder. Each kind of work — browsing the web, emailing people outside, publishing, deleting, spending credits, running commands on a computer — gets a rung: Off, Ask or Auto, plus Draft where there is something to read first, such as an email or a command. Set a rung for the workspace, then narrow it for one Agent. Emailing people outside, publishing and deleting never go above Ask, and moving money stays off.
- What the rails stopped. A log of the actions the platform refused, naming the rule, the Agent and the kind of work, kept for 90 days.
- Readiness hints point out when a kind of work looks ready for more trust. Nothing is promoted automatically.
The checks run in the platform before an action happens, so an instruction or a document cannot talk an Agent past them. Only a signed-in owner can change a rung.
Limits: the ladder exists today through the API only. Covering deleting, spending, access changes and computer commands is part of this step. Approving a stopped action so it runs exactly as prepared comes later, and until then only rungs you set yourself are enforced: the defaults are shown but not applied.
Share a live, read-only view of your Task board with one link
Settings → Sharing will give an Organization's owner one link that shows the Task board, your Agents and recent activity to anyone without an account, and stops working as soon as you regenerate it.
What is coming: a Sharing tab in Settings and a public page behind the link.
- One read-only link. Visitors see the Task board updating live, your Agents and their status, and recent activity. They never see chat, decisions, runs, costs, email, memory, settings or anyone's name.
- You stay in control. Preview the page as a visitor, regenerate the link to cut off everyone who has the old one, or turn sharing off. Settings shows how often the link was opened, and you are notified the first time someone opens it.
- Search engines are asked to stay away unless you allow indexing.
Only the owner can see the link or change sharing. Other members see the settings read-only.
Limits: sharing belongs to an Organization, with one link each. The platform side behind the link exists today, but there is no screen to turn it on yet. Publishing chosen Knowledge Library documents through the same link is a later step.
Download a complete backup of your workspace
Settings → Data will get a backup card that builds one dated .zip of everything Ever Works holds for your workspace, with a manifest of what is inside. Credentials are noted, never copied into it.
What is coming: a backup card at the top of Settings → Data. The existing export, import and GitHub sync stay as they are, below it.
- One archive. A dated
.zipcovering fifteen areas of your workspace, from Agents, Tasks and Works to runs, schedules and connections, plus the files you uploaded. - A record of what is inside. A manifest lists what landed, what was trimmed and why, and what a future restore could put back, with a plain-language README and a checksum for every file.
- No secrets. Keys, passwords and tokens are left out. The archive only notes that a credential was set, so you know what to enter again.
- History. Follow progress, cancel a running backup, and download or delete earlier ones. The Danger Zone shows when you last backed up.
Only the workspace owner can create or download a backup. By default you can make 3 in any 24 hours, and each archive is kept for 14 days.
Limits: the engine behind this card is merged but not yet working end to end, and corrections are in review, so nothing is switched on for you yet. An archive is not encrypted, so keep it somewhere safe. Restoring from one is a later step.
- One archive. A dated
See why an Agent is paused or not picking up work
Each Agent's page will state a plain reason next to its status, such as paused by you, blocked on a credential or stopped after repeated failures, with who paused it, when, and the work waiting for Resume.
What is coming: a status card at the top of each Agent's page, and a status dot with a short reason on the Agents list.
- A reason, not just a colour. Working, idle, waiting on you, paused by you, blocked on a credential, stopped after repeated failures, stopped by the platform, not started or archived, with a link to the failing run, the connection to fix or the decisions waiting for you.
- A clearer Pause. Add an optional note saying why, see which runs are still finishing, and choose whether to stop one now.
- Held work. A list of the work the platform held for the Agent while it was paused. Resume releases it oldest first and tells you how many items went out.
Limits: the reasons, the pause note and the held-work list already exist in the platform, but no screen shows them yet. Levels and a written personality for each Agent are later steps.
See where your credits went, by kind of call and Mission
Billing and Usage & Credits will split spending into your own model keys, credits and add-ons, break credits down by kind of call and by Mission, and add a credit price list page.
What is coming: new cards and panels on the Billing and Usage & Credits pages.
- Three meters that never overlap. Model usage paid with your own provider keys, credits paid through Ever Works, and add-ons such as email sends.
- Where the credits went. The top kinds of call (such as web search) and the top Missions for the period, with an exact remainder, work outside any Mission on its own row, and each row linking to the matching runs.
- A credit price list. What one web search, page extraction or screenshot costs in credits, with the version of the list.
- Run receipts itemise cost by kind of call, including calls that were cached or failed.
Already in place: every new usage record says which meter it belongs to, who paid, and whether the call succeeded, was cached or failed. Failed calls are recorded at zero credits.
Limits: your bill does not change. By default credits are still charged from the provider's cost, and the price list page will say whether its prices are what your deployment charges. Spending caps that stop work are a later step.