Included AI lets Sentinel run updates for a site without an AI key of your own. It may not be available to every organisation yet. Where it is switched off, you do not see it and you choose your own key. Sentinel never names the model or the provider.
Choosing it. In the enrolment wizard, the AI step is titled "Which AI should run updates for this site?". The choices come in this order: Included AI (marked Recommended, no key needed), then your organisation's own keys, then Add a new key. Included AI says why it is recommended: its models and skills are tuned and measured together. If your organisation has no AI key, Included AI is chosen for you. Choosing Included AI, like choosing a key, needs the Patching AI acceptance on that step (see Enrolling a site). Only organisation administrators can add Included AI, with a recent second-factor code if they have set one up. Adding it creates an Included AI connection under Connections. It becomes the default only if you have no default key. Included AI is not available to organisations covered by a Black membership, or by Agency on the new prices; they use their own key. A platform administrator can switch Included AI off for an organisation. Its Included AI jobs then stop, you must choose your own key, and adding Included AI again is refused.
Limits for each site. These limits are for the previous prices. On the new prices there are no monthly counts of security or maintenance updates: the one monthly limit is the share of the site's Included AI that is used, shown as a percentage (see On the site, under the newer Pay as you go pricing below).
- Security updates: up to 4 a calendar month, in the organisation's time zone. The limit is fixed: a Critical or Highly critical advisory does not add one. A fifth update waits for the 1st of next month, and you can switch the site to your own key at any time.
- Maintenance updates: 1 a month.
- A job counts once it has called the AI or opened a pull request. Jobs that found nothing to update, were cancelled before they ran, or failed because of Sentinel do not count. Rebase and CI fix jobs do not count as jobs.
- CI fix rounds are limited to 1.
- There is also a monthly estimated-spend limit for each website, in US dollars. Billing shows the amount before you buy, the wizard shows it before you accept Included AI, and the site's card shows what is used and what is left. It resets at the start of each UTC calendar month and can stop work before the monthly updates are used. When it is reached, new Included AI work waits until next month.
Included AI is part of the plan at no extra charge. It is allowed during a legacy Sentinel trial, with the same limits as above. On newer Pay as you go accounts, work beyond the monthly Included AI can use AI credit (see Billing) where it is open to you. The Pay as you go free first site can use it with the same limits; see Patching on your free site under Billing. Autopilot uses the same limits.
Schedule. With Included AI, maintenance can only run monthly or quarterly. Weekly and fortnightly need your own key, and saving them is refused. See Schedules.
On the site. The site's Patching tab shows an Included AI card with "Security updates this month: X of Y" and "Maintenance in (month): X of Y"; both reset on the 1st. It also shows "Estimated spend this month" against the limit and how much is left. When a security update is waiting, the card also says "Waiting: monthly limit reached", shows when it runs next, and offers Use your own AI key for unlimited updates. The card also says what is sent: the files the update works on and its command output go to Sentinel's AI provider, under Sentinel's account, only to make the update, and the provider does not train on them.
On the site, under the newer Pay as you go pricing. Where that pricing is open to your organisation, the card is the AI panel and shows percentages, not counts. It has four states:
- Within Included AI. The heading reads "Included AI, (month)". It shows how much of the month's Included AI is used as a percent, "Resets on 1 (month)", and every job this month with its share as a percent, largest first. The shares add up to the percent used, apart from rounding and security scans, which count but are not listed. The job list shows to members who can see Patching jobs. If something else stops Included AI, such as a platform administrator switching it off, the card shows that message. It ends with "Want no limit? Add your own OpenAI or Anthropic key. The site price stays the same." The link opens AI keys in Patching settings. Add the key there, then choose it for this site on the Configuration tab.
- Paused at 100%. The card says "New patching jobs for this site are paused until (date)." If a security advisory is waiting, it adds "A security advisory is waiting:" and the project name. It offers three choices. Add your own key links to AI keys in Patching settings; add the key there, then choose it for this site on the Configuration tab. Buy AI credit shows pack prices and a Buy credit button to a billing administrator; everyone else reads "A billing administrator can buy AI credit for this billing account on the Billing page." Wait says "Included AI resets on (date). Waiting security updates run then; other updates run at their next scheduled time."
- Using AI credit. Past 100%, with credit on the billing account, the card shows Using AI credit. Administrators who can read billing also see "AI credit left for this billing account" in money and the auto recharge line. Billing administrators also get Buy credit and Set up auto recharge (or Change auto recharge). Anyone else reads that the site carries on with the billing account's AI credit, with no amounts. When the credit runs out, patching pauses again.
- Your own key, Agency or a membership. The heading reads "AI for patching" with a Your key badge. It says "This site uses your (provider) key ending (last four characters). You pay (provider) directly. Sentinel charges no AI. Your site price stays the same: the Included AI is not refunded." Agency sites and memberships add "Agency sites and memberships always use your own key." This state shows even when Included AI is not offered to your organisation. A link opens Patching settings for the key's spend limits. If the site has no working AI key yet, or the key chosen for it was revoked, the card is not shown and the setup steps tell you what to do. Every Included AI state ends with the note that the files the update works on and its command output go to our AI provider, which does not train on them.
Any member who can see the site sees the percent and the job shares. Money is for administrators who can read billing, and buying is for billing administrators. The server enforces this, so a hidden button is not the only protection.
Messages. Organisation administrators get one email and one message in Messages a month for each site when a scheduled security update is waiting for the limit. It says when the update runs and that the site can switch to its own key now.
Home. The Patching section of Home shows a line such as "Included AI: N sites. M at its monthly security limit", with links to those sites.
Pages under Patching
- Home (
/app/home): a Patching section for people with Patching read, while Patching is on for your organisation. It shows four counts (Sites enrolled, Ready to update, On Autopilot and Deployed this month) and one usage card with the Included AI line. With no site enrolled it says "No site is enrolled in Patching yet. Enrol a site from Sites". Waiting on you and Recent activity are on Home too. There is no separate Patching Overview page any more. See Patching jobs and activity. - Sites: there is no separate Patching Sites page. The Patching column, the Patching filter and the Enrol link on Sites do the job. Sites that need a person show their reason in the Status column.
- Jobs (
/app/jobs): every update job of your organisation, with state chips, filters and a drawer for each job. Links above the page, Jobs and Every step, switch between the two views. It is empty until the first update job runs. The old address/app/patching/jobsno longer opens. See Patching jobs and activity. - Every step (
/app/jobs?view=activity): a second view of Jobs, with every recorded step across your sites. There is no separate Activity page, and the old/app/patching/activityaddress no longer opens. Each site's own activity is also on the site's Patching tab. - Deployments (
/app/deployments): every deployment of your organisation, with its own page for each one. - Settings (Settings → Patching): three pages, Connections, AI keys and Profiles, listed under the Patching heading in the Settings menu. The old address
/app/patching/settingsredirects there.
Connections
Connections are things you connect once for the whole organisation. The Connections page (Settings → Patching) lists Git hosts and hosting platforms, with a card that points to the AI keys page:
- Git hosts: GitHub.com and GitLab.com accounts, shown to administrators only. Connect each GitHub organisation separately.
- Hosting platforms: the accounts where your sites run.
Below them, administrators also see GitHub identity for override merges. Select Connect my GitHub account and approve the request on GitHub. The connection can only do what both the Sentinel GitHub App and you can do, and only where the App is installed. Sentinel stores the token securely, never shows it, and uses it only to merge a pull request on your instruction when its approval is missing, the Sentinel App cannot bypass the rule and the branch protection is classic and not enforced for administrators. You must administer the repository. Each account is listed with when it was connected and when it was last used; Remove (after confirming) deletes it. If your deployment has no GitHub sign-in application set up, the section says so.
Git hosts and hosting platforms can be connected now (see below). You can read the list, rename a connection and Remove it (after confirming). AI keys are managed on their own tab (see AI keys). Keys and tokens are never shown after they are saved.
Remove is refused while a site's setup uses the connection. The message lists those sites, each linking to its Patching tab. Change the setup there, then remove the connection. If your repository or platform is not offered, use My repository is not listed or My platform is not listed.
AI keys
AI keys are your own API keys for an AI provider. Sentinel does not bill you for AI use: the provider bills your own key directly. Included AI is the alternative, and it is part of the plan price. Each key's cap is a stop inside Sentinel, based on an estimate. It is not a ceiling on your provider's invoice. Open Settings → Patching → AI keys. Organisation administrators manage keys; other members read the list only. Amounts are in US dollars.
The table shows, for each key:
- Name, with a Default badge on the organisation's default key and its description.
- Provider.
- Model: the model and reasoning level the key's jobs run on, for example "Claude Opus 5.5 at a high reasoning level", or Platform default when the key has no choice of its own. If Sentinel stops offering a model you chose, the column adds "Paused: chosen model no longer offered". Jobs on that key are paused until an administrator picks a model and selects Save model (choosing Platform default counts), and every organisation administrator receives one email about it.
- Status. A key the provider rejects, rate limits or reports as out of credit shows an error with a Needs attention reason, and the sites using it show the AI key step as not done. Every organisation administrator receives one email per key per day ("Your provider connection needs attention").
- Spend cap: the cap and "$X used this month". Administrators can use Change cap. A key with no cap of its own uses the platform default. Only money already spent counts towards the cap, not jobs that are still running. When another job is running on the same key and could still use up the rest of the cap, the next job waits in the queue (it does not fail) and starts with its AI agent when that job finishes. It keeps its place. Once the spend reaches the cap, new jobs on that key stop starting until next month.
- Used by: the enrolled sites and profiles that use the key.
- Last used.
A line above the table shows this month's estimated spend. The figure is an estimate, not your provider invoice. Each key has its own cap; there is no organisation-wide cap. Once a key's estimate reaches its cap, jobs on that key stop starting, while jobs on other keys keep running.
Add a key. Choose Add a key and fill in Provider, Name (required) and API key. Model for jobs is optional. Its first option is "Platform default" followed by the platform model and its reasoning level in brackets, which is what Sentinel runs when you do not choose. The other options are only the models Sentinel has tested for customer keys. Choosing one shows Reasoning level, with the levels offered for that model. Select Test connection: Sentinel checks that the key is valid and can reach the model it will run on, and the model list then narrows to what the key can reach. If the key is valid but cannot use the model you chose (or the platform default), the test and saving are refused with a message that names the model and asks you to check the key's project or model access at the provider. Fix the access at your provider, then test again, or choose another model. Nothing is kept until you save. Optionally set a monthly spend cap in US dollars and a description, and tick Make this the organisation's default key if you want it. The key is write-only and is never shown again. Saving or rotating a key needs a recent second-factor check, so you may be asked to confirm it again first.
What you can choose. Today the list is OpenAI GPT-6.1 Sol at a high or xhigh reasoning level, and Anthropic Claude Opus 5.5 at a medium, high or xhigh reasoning level. Nothing else is offered: no other models, no other providers and no custom endpoint. Prompts, skills, tools and budgets cannot be changed. With no choice, an OpenAI key runs GPT-6.1 Sol at a high reasoning level and an Anthropic key runs Claude Opus 5.5 at a medium reasoning level. Included AI always runs the platform choice and cannot be changed.
Manage. Select Manage on a key to rename it, change its description, change its model, run Verify key again (it also checks that the key can reach the model), or Rotate key to paste a new key. In the model section, pick a Model (Platform default or a model the key can reach), pick a Reasoning level, and select Save model. Setting both back to the platform default returns the key to the platform choice. If you choose a model or level that is not offered, Sentinel refuses it ("ai_model_not_offered") with a message that lists what that provider's keys can run on. If the key cannot reach the model you chose, saving is refused ("ai_model_unreachable") and nothing is stored. Choose both a model and a level, or neither. Make default makes a key the one that new enrolments preselect.
Revoke a key. Select Revoke.
- A key nothing uses asks for a plain confirmation. The key is deleted from the vault and cannot be restored.
- A key that is in use shows a warning with how many sites and profiles use it, and two choices. Move them to another key, then revoke moves the sites and profiles to a replacement key you select (or choose Or add a new key here) and then revokes the old one. Revoke without a replacement needs you to tick "I am aware that this key is in use and revoking it will break integrations" and confirm. Those sites then show Needs attention and stop running updates, and profiles that use the key cannot be used until they have a key.
Your choice is recorded in the activity log. If a site still names a revoked key, saving or enrolling it says the key was revoked and asks you to choose another key.
On a site's Patching tab, AI key shows which key the site uses. Administrators can pick a different key there. When the site uses your own key, the Configuration summary also has an AI model line naming the model that key runs, either its own choice or "Platform default" with the model. Included AI shows no model line.
Reviewing changes
When Patching writes a change, a separate AI agent reviews it before the pull request is opened. The site's own AI key still writes the change. This part may not be switched on for your organisation yet. Where it is off, you do not see these choices.
- Organisation default. On AI keys, the Reviewing changes section sets which key reviews for every site. The choices are Default: an Anthropic key reviews when an OpenAI key codes, The coding key reviews too, as a separate agent, or one of your active keys by name. The choices appear only when your organisation has keys from both Anthropic and OpenAI. With keys from one provider, the section says that a second agent reviews on the same key. When Included AI is not on for your organisation, the section also shows Recommended: Included AI, because Sentinel tunes and measures its models and skills together.
- One site. On the site's Patching tab, Which AI key reviews the change has Follow the organisation as its first option, then the same choices. It is not shown when the site uses Included AI.
- Who can change it. Any member can read the organisation default. Changing it needs an organisation administrator. Changing a site's choice needs the same permission as choosing the site's AI key, which is permission to run Patching updates.
- Model and cost. Each key runs on the model chosen for it, and its provider bills it. The reviewer's use is counted against the reviewer's key and its monthly cap, and shows as its own row on the AI keys spend list.
- When the chosen key cannot be used. If the key was removed, is not active, has no accepted Patching AI terms for the site, is busy with another job, or has reached its cap, the review runs on the coding key. The job's activity log says "The reviewer runs on the coder's AI key for this job".
