Settings, Permissions & Your Data
Switch the AI Assistant on for your organization, choose what it may use and which models members may pick, see what each model costs and how paid models are billed, set a budget, decide who may use and manage it, and know where conversations are processed.
On this page
The assistant is off until an organization admin switches it on. Everything an admin controls lives under Configure Settings → AI Assistants. The Agent Fleet, which shows your organization's AI agents at work, is in the Agents module, not here.
Turn the assistant on or off#
Open Configure Settings → AI Assistants → Settings:
| Setting | What it does |
|---|---|
| Enable the AI Assistant for this organization | On shows the assistant button to every member whose role allows it. Off hides it for everyone, and nothing is sent to any model provider. |
| What the assistant may use | Choose the sources the assistant may draw on: Documentation (devstride.docs.devstride.com), DevStride data (your items, boards and people: the assistant can read them and propose changes, each one approved by the member) and GitHub (the repositories your organization's DevStride GitHub App installation can reach: the assistant can list them, search their code, read a file and find an item's pull requests; it only reads, and nothing is written to GitHub). Every member who can use the assistant can read what the installation reaches, whatever their own GitHub access, so switch GitHub off if that is too wide. With no GitHub account connected the setting says there is nothing to read yet; see GitHub to install the app. |
| Models members may choose | Limit which AI models members can pick. Leave it empty to allow every free model. A paid model (marked paid in the list) is offered to members only once you add it here, and a list that holds paid models must keep at least one free model on it, so members always have a model to fall back to. Members see a model picker only when more than one model is allowed. |
Changes apply as soon as you switch them. A member with the assistant open sees the button disappear the next time the page loads.
Prices, and paid models#
Configure Settings → AI Assistants → Prices lists every AI model DevStride offers with what it costs your organization, in US dollars per million input, output and cached input tokens, and whether each model is on your allow-list. A model marked Free for now costs nothing until DevStride ends its introductory period; the page says so, and the staff who end it see the prices before they do.
How paid models are billed:
- AI usage is billed monthly, in arrears, on your existing bill: the month's paid-model usage becomes one line, named by month, on your next card charge (the receipt and Billing History show it) or on your next QuickBooks invoice. The amount is the provider's cost with DevStride's margin included, exactly as the Prices page shows.
- Only usage made while your workspace was billable is charged. A free trial is never billed for AI usage, including the days of a month before the trial converted; a demo, copied or trial workspace, or one DevStride has waived, is never billed.
- A workspace included in another workspace's license has its usage billed on that license's bill.
- Paid models are offered only while DevStride can collect for them. A workspace with a failed card payment or an overdue invoice falls back to free models until the payment is settled; a member who asks for a paid model meanwhile gets the answer from the free default, and the conversation says so. Workspaces on an annual plan cannot use paid models yet.
The page reads This workspace is not billed for AI usage at the moment when none of your usage is charged (on trial, included in a license, or waived). Anyone who can open Settings and holds Use Ai Assistant, Manage Ai Assistant or View Ai Usage can see it; Your Plan links to it too.
See what the assistant costs#
Configure Settings → AI Assistants → Usage shows this month's AI Assistant cost for your organization: the total so far, a projection for the whole month at the current pace, and the breakdown by model and by member. Costs are shown in dollars at the price DevStride charges for each model. People with View Ai Usage see this page; Manage Ai Assistant includes it.
Set a budget#
Configure Settings → AI Assistants → Budgets lets you cap what the assistant may spend each month:
| Setting | What it does |
|---|---|
| Organization budget | The most the whole organization may spend in a calendar month. Leave it on No limit to use DevStride's default budget, which the page shows as the inherited value. |
| Budget per member | An optional cap for each member within the organization budget. It can never exceed the organization budget. |
The page shows the month so far against the budget; the bar turns orange at 80% and red at 100%.
What members experience: when an answer takes a budget past 80%, the assistant adds a notice to that answer saying so. At 100% the assistant stops before it would spend anything more, and the notice names whose budget ran out (the member's or the organization's) and who can raise it; members who may manage the assistant get a link straight to the Budgets page. A raised budget takes effect on the next question. A budget is checked at the start of a run and again before each step, so a long answer is cut off rather than overrunning. Budgets reset at the start of each month.
DevStride also keeps a limit of its own on total assistant spend across every organization; if that is ever reached, the assistant pauses for everyone until the next month, and the notice says so.
Connect an external AI client#
The same AI Assistants section holds Connect AI: the MCP Server URL and ready-to-paste setup snippets for Claude, ChatGPT, Cursor, Codex, Copilot and Gemini CLI, so an agent you run elsewhere can work in DevStride as you. The page is shown to AI Assistant admins; any member can still connect by following Setting Up Your Connection, which gives the same URL and steps.
See what your AI agents are doing#
Agents → Fleet, in the Agents module, shows what your organization's agents are working on right now, and it updates live. Every Claude Code session that runs the DevStride plugin (version 4.0.0 or later) appears there by itself, whoever in your organization starts it and whether or not it runs inside a code repository: the plugin reports when the session starts, while it works, while it waits on its person and when it ends. The only thing a machine needs is a DevStride connection made with an API key (see Setting Up Your Connection); a machine connected by browser sign-in alone is not reported. You see only your own organization's agents.
One row of tiles across the top sums up the fleet, including how many sessions are waiting on a person. When the row is wider than the page it scrolls sideways (trackpad, Shift with the mouse wheel, touch, or Tab through the tiles). Below them, each agent has one row: its name, what it is working on (with links to those items), its status and when it was last seen. A live session's status says what it is doing: Working, Waiting on a person (it has finished its turn, asked a question or is waiting for a permission) or Starting. Expand a row for the details: session id, addresses, branch, plugin version and start time. All times are on your own clock. When none are working, the page says so.
When an agent session stops#
When DevStride's agent session health check is turned on, the page also tells you when a session goes quiet. Each session reports a status ping at a regular interval:
- Warning (yellow): the session missed a status ping. The row says how many it missed and how many more tries are left.
- Stopped (red): it missed five pings in a row, or it ended while it still held work.
- Stalled (orange): it says it is working but has done nothing for ten minutes, or for sixty minutes when it is in the middle of one long step such as a test run. A session that is waiting on a person is never marked Stalled.
- Recovered: a session that starts reporting again turns green.
A session that ends normally leaves the page. A session that holds no work and goes quiet still shows Warning and then Stopped, marked as holding no work, and nobody is notified.
Agents → Log lists every one of these changes, newest first: when it happened, which agent, what happened, what work it held and whether anyone was told. Select an agent's name to show only that session.
When a session stops or stalls while it holds work, the person whose key launched it gets one notification, following their own notification settings. The notification opens the log at that session. A stall is reported once per session for each piece of work: if the same session stalls again on the same work, the new row appears on the page and in the log, marked Not sent again: already told about this, and no second notification goes out. A stop is still reported every time it happens. The check only reports: it never releases or takes over a session's work.
To open the Agents module you need the View Agents Module permission. The Owner and the Admin role hold it, and so does every role that already held Manage Agent Claims; add it to any other role in Roles & Permissions. Anyone with it can see the session details behind each row, including addresses, so grant it with that in mind. The module's Workflows view, which shows the step each running agent is on, is described in Agents Overview. A member needs no permission for their own sessions to appear on the page.
Who can do what#
Three permissions in the AI Assistant group of Roles & Permissions govern the assistant:
| Permission | What it grants | Member | Admin |
|---|---|---|---|
| Use Ai Assistant | Open the AI Assistant and chat with it. | yes | yes |
| Manage Ai Assistant | Switch the assistant on or off, choose which sources and models it may use, set the organization's Budgets, see Usage and Prices, and open Connect AI. Requires the Settings module. | no | yes |
| View Ai Usage | Open Configure Settings → AI Assistants → Usage (this month's AI Assistant cost for the organization, by model and by member) and Prices. Requires the Settings module. | no | yes |
Every role that existed when the assistant launched received Use Ai Assistant. Remove it from a role to hide the assistant from that role's members while leaving it on for everyone else.
Beyond these three, the assistant has no permissions of its own. It reads and changes records as the member who is chatting, so your existing roles and sharing decide what it can reach.
Your data#
Conversations are processed by OpenAI, L.L.C. and Anthropic, PBC as DevStride sub-processors. The text of a message, and the context your organization allows, go to the chosen model's provider only while a member is chatting. Nothing is sent while the assistant is switched off.
Conversations are private to the member who had them. Changes the assistant makes appear in each item's Activity as that member's, marked via AI Assistant.